Survival and recurrence after first cerebral infarction

Stroke is a leading cause of morbidity and mortality, afflicting more than 500,000 Americans each year.¹ Consequently, precise information on rates, temporal trends, and determinants of survival and recurrence after first stroke is needed to design clinical trials as well as to inform decision models^2,3 pertaining to health care policy research into primary and secondary stroke prevention. Despite numerous studies, considerable uncertainty exists about which putative risk factors are independent determinants of survival and recurrent stroke after a first stroke.^4-15 Moreover, although previous population-based studies demonstrated progressive improvement in survival after stroke during the 3 decades after 1950,^4,16-18 a recent study suggests that the rate of this decline is decreasing,¹⁹ and yet another suggests that it is continuing.²⁰ To address these issues, we examined more recent temporal trends and used new statistical validation methods to model determinants of survival and recurrent stroke in the cohort of all residents of Rochester, MN, who sustained a first cerebral infarction from January 1, 1975, through December 31, 1989.

Methods. The Rochester Epidemiology Project medical records linkage system provides resources to identify nearly all new cases of stroke in the community.²¹ Virtually all medical care in the community is supplied by the Mayo Clinic and its two affiliated hospitals or by Olmsted Medical Group, a smaller group practice, and its hospital. In these institutions, all medical diagnoses made for a resident of Rochester are entered on a master sheet in the patient's medical record, which is then entered into a central computer file. The computer file includes diagnoses made for Rochester residents at other medical practices in surrounding communities, the University of Minnesota, and the Veterans Administration Hospital in Minneapolis. This file provides access to all inpatient and outpatient data, emergency room visits, nursing home care, and autopsy or death certificate information. Population-based studies of stroke in Rochester are approved by the Mayo Foundation Institutional Review Board.

The medical records of all residents of Rochester who had a diagnosis of stroke, transient ischemic attack, or disorders that could be mistaken for stroke or transient ischemic attack during the 15-year period from January 1, 1975, through December 31, 1989, were screened by a neurologist and a trained nurse abstractor to determine whether the case met the criteria for stroke. All determined cases then had verification of residence based on information from city and county directories and earlier medical records. The type of stroke was also determined, with imaging studies and autopsy reports used when available. Death certificates and autopsy protocols were also reviewed to identify those with the diagnosis of stroke. The clinical record was then reviewed to determine whether any clinical symptoms were consistent with stroke. The categories of first stroke are defined in theAppendix.

To exclude persons who may have moved to Rochester to facilitate treatment of diagnosis of an existing disorder, we restricted eligibility to those who had lived in Rochester for at least 1 year before the stroke. Patients with a clinical diagnosis of stroke or with stroke listed as a cause of death on the death certificate who died within 24 hours of symptom onset were excluded if there was no clinical evidence of a focal neurologic deficit, no CT or MRI studies done, or no autopsy performed. Head CT, head MRI, or brain autopsy was performed in 92% of cases in the 1985-1989 1989 quinquennium, 88% in 1980 to 1984, and 73% in 1975 to 1979.²²

Using the described diagnostic rubrics and ascertainment procedures, we identified 1,111 residents of Rochester, MN, who had a first cerebral infarction during the 15-year period of the study. The neurologist and nurse abstractor abstracted the medical record of each patient in this cohort and recorded on standardized forms information on stroke risk factors before or at the time of first cerebral infarction, treatment before stroke, date of last follow up or date of death, cause of death, and date and type of stroke after cerebral infarction, all prior to January 1, 1993. A partial list of variables appears in table 1, and definitions for the variables studied have been published elsewhere.¹⁰ Stroke was coded as the cause of death only for patients who died from a primary neurologic process attributable to stroke (brain herniation, for example).

Recurrent stroke was defined as a new neurologic deficit that fit the definitions for the stroke subtypes outlined above, occurred after a period of unequivocal neurologic stability or improvement lasting at least 24 hours, and was not attributable to edema, mass effect, brain shift syndrome, or hemorrhagic transformation of the incident cerebral infarction. Autopsy documentation of recent infarction in a vascular territory different from that of the incident infarction was counted as a recurrence if the date of the recurrent stroke could be estimated.

Statistical analysis. We used the Kaplan-Meier product limit method²³ to estimate rates of survival and recurrent stroke after first cerebral infarction. In both the survival and the recurrence analyses, we censored patients at the time of migration from Olmsted County, at the time of last follow-up, or on January 1, 1993. In the recurrence analysis, we also censored, at the time of death, patients free of second stroke. The one-sample logrank test was used to compare survival estimates to expected survival based on the Minnesota white population age- and sex-specific mortality rates and recurrence estimates to expected values based on the stroke incidence rates of the Rochester population.

We used Cox proportional hazards regression analyses with bootstrap validation^24,25 to model the role of purported risk factors in survival and stroke recurrence after first cerebral infarction in the following fashion. Univariate Cox proportional hazards regression analyses were used to estimate the risk ratios for all factors under consideration. We examined Schoenfeld residuals²⁶ to assess the proportional hazards assumption for each of the factors. Then multivariate proportional hazards regression analyses were performed to identify the collection of risk factors making independent contributions to survival and risk of recurrence. This was done by stepwise forward and backward variable selection procedures, involving not only the basic variables but also their interactions. The collection of risk factors and interactions that resulted was validated by bootstrap resampling methods.²⁵ Bootstrapping is a technique in which 500 multivariate proportional hazards regression analyses are performed from 500 data sets generated by random resampling from the original data set. Variables are included in the final model if they are validated in over 70% of these analyses. We examined temporal trends in survival and recurrence after first cerebral infarction using year of first cerebral infarction and the three quinquennia of the period of the study (1975 to 1979, 1980 to 1984, 1985 to 1989) as variables in the univariate and multivariate settings. The effect of recurrent stroke on survival after first cerebral infarction was examined by use of recurrent stroke as a time-dependent variable after adjustment for determinants of survival identified in the final model.

Results. During the 1975 to 1989 period, 1,111 residents of Rochester, MN, had a first cerebral infarction. A total of 649 (58%) were women. Mean age was 74.5 (±13.0) years. A total of 911 patients (82%) were hospitalized for the first cerebral infarction. A total of 888 (80%) had head CT (or MRI) or a brain autopsy, or both.

Survival. A total of 764 patients (68.8%) died during the period of follow up after first cerebral infarction.Table 2 presents the Kaplan-Meier estimates of probability of survival after first cerebral infarction. Survival after first cerebral infarction in these patients was significantly lower in comparison with estimates of expected survival derived from the age- and sex-specific mortality rates of the Minnesota white population during the period of the study (one-sample logrank test, p < 0.001)(figure 1). Even

Table 3 presents distributions of cause of death during each of the 3 quinquennia and beyond. Stroke as a cause of death subsequent to first cerebral infarction declined over the period of the study, whereas deaths due to other causes (such as pulmonary embolism or pneumonia) increased. The percentages of deaths caused by myocardial infarction, congestive heart failure, and sudden unexplained death remained relatively unchanged over the period of the study.

Determinants of survival. The univariate Cox proportional hazards regression analyses of possible determinants of survival after first cerebral infarction are presented in table 1. A multivariate Cox proportional hazards model was developed to determine which of the variables displayed in table 1 had independent influence on survival. The results of the multivariate analysis are summarized in table 4. Although 16 variables appeared to be significant predictors of survival in the univariate analysis, the multivariate model indicated that ischemic heart disease (in younger patients; figure 2) and congestive heart failure(figure 3) were the most important predictors of death after first cerebral infarction, followed by age and persistent atrial fibrillation (figure 4). The contribution of ischemic heart disease to the risk of death after first cerebral infarction became increasingly less important with increasing age (seetable 4 and figure 2). Adjusting for the variables in the multivariate and proportional hazards model, recurrent stroke as a time-dependent variable was a strong predictor of death after first cerebral infarction (risk ratio [RR], 3.0; 95% confidence interval [CI], 2.5 to 3.5; p < 0.0001).

Cox proportional hazards regression analysis was used to assess the effect of the date of stroke on rate of overall survival. Whether expressed as the year or the quinquennium of the first cerebral infarction, there was no significant temporal trend in survival in either the univariate or the multivariate model.

Recurrence. A total of 272 patients (24.5%) had a subsequent stroke after first cerebral infarction. Most recurrent strokes were cerebral infarcts (248, 91.2%). Table 2 presents the Kaplan-Meier estimates of the probability of subsequent stroke after first cerebral infarction. The probability of subsequent stroke after a first cerebral infarction was significantly greater than the expected probability of stroke based on age- and sex-specific stroke incidence rates in the Rochester population during the period of the study (one-sample logrank test, p < 0.001) (figure 5).

Determinants of recurrence. The univariate Cox proportional hazards regression analyses of possible determinants of survival free of subsequent stroke (recurrence) after first cerebral infarction are presented in table 1. A multivariate Cox proportional hazards model was developed to determine which of the variables displayed intable 1 had an independent effect on recurrence. Univariate analyses suggested that 12 risk factors were predictors of stroke recurrence after first cerebral infarction, but in the final multivariate model, only age (RR per decade, 1.2; 95% CI, 1.10 to 1.36;p = 0.0002) and diabetes mellitus (RR, 1.7; 95% CI, 1.26 to 2.24; p = 0.0004; figure 6) were significant independent predictors of recurrence. Cox proportional hazards regression analysis demonstrated no significant temporal trends in the rate of stroke recurrence during the 15-year period of the study.

Discussion. Our study provides new information on temporal trends in survival after first cerebral infarction that may have important implications for health care policy research. The trend toward improvement in survival after first cerebral infarction documented by us and others during the 3 decades after 1945 has ended. In a previous population-based study in Rochester, MN, we found improvement in both long-term and short-term survival between 1945-1949 and 1975-1979.¹⁶ This trend was also noted in Framingham,¹⁷ in Japan,¹⁸ and in rural North Carolina.⁴ The trend toward improvement in survival after cerebral infarction over time had been attributed to improved management of cardiac disease, improved treatment of other complications after cerebral infarction, increased utilization of rehabilitation services, detection of milder cases of cerebral infarction, or some combination thereof.^16,27

An examination of the cause of death (see table 3) in our cohort may partly explain why survival after first cerebral infarction did not continue to decline over the 1975-1989 period. We found that death attributable to stroke after first cerebral infarction declined, but this trend was offset by an increase in death after cerebral infarction due to noncardiac and nonstroke causes (see table 3). This finding may indicate that cerebral infarction has become less lethal but that the disabled survivors are more likely to succumb to pulmonary embolism, pneumonia, or other non-neurologic afflictions.^5,17 Death after cerebral infarction attributable to cardiac causes (including unexplained sudden death) remained relatively stable despite advances in management of cardiac disease over the time of the study.

Our results are substantially different from those in a recent study from the Minnesota Stroke Survey, which reported improvement in survival after cerebral infarction during the 1980s.²⁰ Shahar et al.²⁰ studied survival after stroke occurring during 1980, 1985, and 1990 in persons hospitalized in the Minneapolis-St. Paul metropolitan area. They found that the odds of death within 2 years after stroke were approximately 40% lower in 1990 than in 1980 and that improved survival after cerebral infarction, especially noncardioembolic stroke, accounted for most of this trend. In contrast, over the same period, we found no such trend in Rochester, MN, a community less than 100 miles from the population center studied in the Minnesota Stroke Survey. Although it is possible that differences in age, sex, race, and risk-factor distributions between the populations of Rochester, MN, and Minneapolis-St. Paul could account for the different findings, we believe that methodologic factors are a more likely explanation. Ours is a study of all individuals in a community with a first cerebral infarction identified through resources of the Rochester Epidemiology Project, regardless of age and whether they were hospitalized for the cerebral infarction. In contrast, the Minnesota Stroke Survey is a study of only those patients in the community who were aged 30 to 74, hospitalized, and identified by International Classification of Diseases(ICD)-9 codes for cerebrovascular disease and whose index stroke may or may not have been the first stroke. Inclusion of nonincidence strokes in a study of the natural history of stroke may affect inferences made about stroke survival, since patients with a stroke preceding the index stroke may have had characteristics that either worsen or improve survival, depending on the nature of the previous stroke or strokes. ICD-9 codes are a less sensitive and specific method of identifying first stroke than the Rochester Epidemiology Project stroke registry.²⁸ Given our finding that age is an important independent determinant of death after first cerebral infarction and given the fact that the population is aging,²⁹ excluding older patients from a study of survival after stroke could result in significant overestimation of survival, which could in turn affect inferences made about temporal trends in survival. Finally, although hospitalization was not an independent determinant of survival in Rochester during the 1975-1989 period, hospital-based studies are subject to selection bias,³⁰ depending on economic, demographic, and social factors specific to a given community that may affect hospital referral and admitting practices in that community. For example, changes in health provision or, as Shahar et al.²⁰ noted in their report, reimbursement (due to introduction of diagnosis-related groups) could have resulted in changes in patient mix or rates of hospitalization for stroke in the Minneapolis-St. Paul metropolitan area during the 1980s. If so, a trend toward increased rates of hospitalization for patients with minor stroke or decreased rates of hospitalization for patients with major stroke in the Minneapolis-St. Paul community could have resulted in an artifactual trend toward improved survival over time.

Survival after first cerebral infarction in our cohort remained significantly lower than in the general population without stroke (seefigure 1). The rates of survival after first cerebral infarction in our study (see table 2) are similar to rates reported from Framingham, Oxfordshire, Perth, and Valle d'Aosta but somewhat worse than those reported from Malmo and northern Manhattan.^6-9,31,32 The worse survival after first cerebral infarction among the largely homogeneous white population of Rochester compared with the ethnically and racially diverse population of northern Manhattan⁹ might be interpreted as suggesting that African Americans and Hispanic Americans have better survival than white Americans after first cerebral infarction. However, we believe that differences in age, sex, medical follow up, and methodology are a more likely explanation for these differences. For example, our study differs significantly from the northern Manhattan study by studying all patients with cerebral infarction in a community, whether hospitalized or not. The northern Manhattan study is of patients in a community who were admitted to the neurologic service of a university teaching hospital. Exclusion of nonhospitalized patients in a study of survival after stroke may introduce bias that could affect survival rates, depending on local admitting and referral practices (see above).

The large number of patients, long period of follow-up, and use of Cox proportional hazards regression analysis with bootstrap validation techniques permitted us to develop more precise models to predict survival and recurrence after first cerebral infarction.

Among 25 variables, many previously identified as risk factors for death after cerebral infarction (see table 1), we arrived at a final multivariate model in which age, congestive heart failure, persistent atrial fibrillation, and ischemic heart disease (especially at younger ages) were significant determinants of survival after first cerebral infarction. In a previous study, largely involving first cerebral infarction occurring in earlier periods in Rochester, we found that age, previous myocardial infarction, atrial fibrillation at stroke onset, congestive heart failure, and an interaction of age and congestive heart failure were the most important predictors of death.¹⁰ Although the models are similar, the present model determined that any ischemic heart disease is a more important predictor of death than myocardial infarction or angina alone and that the interaction of ischemic heart disease with age is more important than the interaction of congestive heart failure with age. The differences in the two studies could be explained in part, by more severe and less successfully treated cardiac comorbidity in the years preceding our current study. However, the new model, unlike our previous model and models derived from other populations, has the advantage of bootstrapping,²⁵ a powerful method of internal validation. Other population-based and nonpopulation-based studies have also identified congestive heart failure,^6,8,9 age,^{4,5,7,9,11-14} atrial fibrillation,^8,12,15 and ischemic heart disease^6,7 as predictors of death after first cerebral infarction. Previously reported independent risk factors for death after cerebral infarction that we considered in our analysis but did not find to be significant independent predictors of death in the final model were hypertension,^4,6,12 gender,^6,11 diabetes mellitus,^4,7,12 and hypertensive heart disease.¹² In addition to differences in statistical methods between our study and other studies, these results may also be attributable, in part, to differences in patient populations, definitions, inclusion criteria, and number of subjects. It is notable that risk factors for the development of first ischemic stroke in our population are not independent predictors of death once the stroke has occurred: male gender, hypertension, previous transient ischemic attack, hypertensive heart disease, and diabetes mellitus.³³ Our final model underscores the importance of cardiac comorbidity as the major determinant of death after first cerebral infarction. Congestive heart failure (RR, 2.3) and ischemic heart disease, especially in persons between the ages of 15 and 60 years (RR, 3.0 and higher), were the strongest predictors of death. Atrial fibrillation was a somewhat less powerful determinant of death but more important than hypertension, diabetes, gender, and preceding transient ischemic attack. Recurrent stroke as a time-dependent variable is an equally important independent determinant of death after first cerebral infarction (RR, 3.0) after adjustment for age, congestive heart failure, persistent atrial fibrillation, and ischemic heart disease.

The high rates of recurrent stroke (see table 2) after first cerebral infarction underscore the importance of investigating secondary stroke risk reduction. Our estimated 5-year recurrence rates (29%) are virtually identical to those reported from Oxfordshire(30%)³⁴ but somewhat higher than those reported from northern Manhattan (22%).⁹ The differences with the last-named study again could suggest that African Americans and Hispanic Americans have lower rates of recurrent stroke than white Americans after first cerebral infarction but more likely reflect differences in age, sex, medical follow-up, and methodology. The recurrence rates for Rochester in the current study are somewhat higher than those previously reported for the 1950 to 1979 period.³⁵ We believe that this difference is related more to study methods than to a change in the natural history of stroke recurrence after first cerebral infarction. During the most recent data collection effort for the 1975-1989 period, increased ascertainment may have been possible because of more complete inclusion of nursing home clinical records in patient histories. Also, for the current study, efforts were intensified to identify stroke recurrence, and histories of patients with first cerebral infarction were reviewed by a trained nurse abstractor and two study neurologists. However, we cannot exclude the possibility that recurrence rates in the 1975-1989 period are actually higher than they were over the decades prior to 1975.

Despite consideration of 22 variables (see table 1), many of which have been previously reported as risk factors for stroke recurrence, Cox proportional hazards modeling with bootstrap validation demonstrated that only age and diabetes mellitus were significant independent predictors of subsequent stroke after first cerebral infarction. Several variables previously identified by some investigators as significantly increasing the risk of recurrence were not found to be independent predictors of recurrence in our final model: hypertension,^9,36-38 valvular heart disease,¹⁰ congestive heart failure,^6,10,38 gender,⁶ previous transient ischemic attack,³⁹ and myocardial infarction or other ischemic heart disease.^6,39 Although atrial fibrillation is a risk factor for first ischemic stroke,⁴⁰ it was not a significant independent predictor of stroke recurrence in our current study and in previous studies from Rochester,¹⁰ northern Manhattan,⁹ or Oxfordshire.¹⁵ In our previous study of cerebral infarction in Rochester, we found that valvular heart disease and congestive heart failure were predictors of recurrence,¹⁰ but these were not significant determinants of recurrence in the final model of the current study. As was the case for the survival analysis, the differences in our current study and the previous studies by us (which included pre-1975 data) and others may be explained, in part, by more severe or less effectively treated cardiac comorbidity in cerebral infarction patients before 1975, or by differences in statistical methods. For example, although congestive heart failure was identified as an independent predictor of recurrence in the proportional hazards regression analysis in the current study, it was not validated in the bootstrap analysis. Other studies identified diabetes^36,37 and increased admission glucose at the time of first cerebral infarction^9,36,37 as independent determinants of recurrence. The identification of diabetes as the single most important predictor (aside from age) of recurrence in our model, and specifically as a more important predictor than hypertension, atrial fibrillation, congestive heart failure, and valvular heart disease, is somewhat surprising. However, this finding supports an emerging body of evidence suggesting that diabetes has ubiquitous effects on the cardiovascular, cerebrovascular, hemorrheologic, and coagulation systems,^41,42 all of which may collectively influence risk of stroke recurrence. In particular, large-vessel^43,44 and small-vessel⁴⁵ intracranial occlusive arterial disease may be overrepresented in diabetics.

Appendix