Stroke Rates During the 1980s
The Minnesota Stroke Survey
Background and Purpose The decline in stroke mortality in the United States may have resulted from declining incidence, improved survival of stroke patients, or both. We previously reported that stroke patients who were 30 to 74 years old and were treated in Minneapolis/St Paul hospitals in 1990 survived longer than did their counterparts in 1980. In the present study, we examined trends in the rate of hospitalized stroke in Minneapolis/St Paul between 1980 and 1990.
Methods For 1980, 1985, and 1990, we obtained lists of discharge codes (International Classification of Diseases, 9th revision) from Minneapolis/St Paul hospitals, identified hospitalizations for acute cerebrovascular disease, and randomly selected 50% of the cases for medical record abstraction. We counted stroke events in five different ways, which were based on discharge codes as well as diagnostic criteria, and computed age-adjusted stroke rates for each year. Stroke mortality in the population was computed for 1960 through 1994.
Results Among men, all five measures of hospitalized stroke attack rate indicated a decline between 1980 and 1985, which ranged from 5% to >20%. Among women, there was a sharp contrast between trends that relied on discharge codes and trends that relied on diagnostic criteria: the former indicated a decline (4% to 19%), whereas the latter indicated some increase. For the second half of the 1980s, most measures of stroke attack rate in men, all measures of stroke attack rate in women, and measures of stroke incidence in both sexes did not indicate a decline in stroke occurrence in the population. Mortality from stroke among 30- to 74-year-old residents of Minneapolis/St Paul, which declined rapidly during the 1970s and early 1980s, declined slowly, if at all, during the second half of the 1980s and early 1990s.
Conclusions The incidence of stroke may have declined among 30- to 74-year-old residents of Minneapolis/St Paul in the early 1980s. However, we found little indication of such a trend between 1985 and 1990, a period of slow decline or no decline in stroke mortality in that population.
Both US national and regional data and data from many other industrialized countries are consistent with a decline in stroke mortality over recent decades.1 2 3 4 5 This trend could have resulted from a decline in the rate of incident stroke, improved survival of stroke patients over time, or both. Understanding the cause of the trend is more than a matter of scientific curiosity: a decline in stroke incidence may be attributed to the success of primary prevention measures, whereas improved survival over time may be attributed to improved acute medical care, improved rehabilitative care, or both. It is also possible, of course, that poorly understood trends in the natural history of stroke have accounted for declining stroke incidence and/or improved survival of stroke patients.
Of the two hypotheses, that of declining stroke incidence has always been more plausible, especially in the era during which much attention has been paid to the detection and treatment of hypertension6 7 and no major advances have been made in stroke-specific therapy.8 Nevertheless, evidence of improved survival of stroke patients has proved to be far more diversified, as reported by several studies for various regions of the United States and for various time periods.9 10 11 12 13 14 15 16 Data compatible with a decline in stroke incidence in the United States are almost exclusively restricted to Olmsted County, Minn, where a decline was noted for the 1950s, 1960s, and 1970s17 but was no longer evident after the late 1970s.18 19
The purpose of this study was to examine trends in stroke occurrence (ie, attack rate and incidence rate) during the 1980s in light of coinciding trends in stroke mortality. We present rates of hospitalized stroke in 1980, 1985, and 1990 in metropolitan Minneapolis/St Paul (the Twin Cities), Minnesota.
The target population consisted of 30- to 74-year-old residents of the seven-county Minneapolis/St Paul metropolitan area: 419 207 men and 451 300 women according to the 1980 census and 550 719 men and 576 690 women according to the 1990 census. We estimated the population size in 1985, by log-linear interpolation, on the basis of its size in 1980 and 1990.
For each surveillance year, we obtained lists of discharge diagnoses from acute care hospitals serving the metropolitan area. For 1980 and 1985, 30 of the 31 area hospitals provided data; the only nonparticipating hospital had a low patient volume. For 1990, all 25 operating hospitals collaborated in this research. From these lists, we constructed a sampling frame for each survey that included Twin Cities residents 30 to 74 years old who had been discharged with one or more of the following acute cerebrovascular disease codes (according to the International Classification of Diseases [ICD], 9th Revision) listed in any position: 431 (intracerebral hemorrhage), 432 (other and unspecified intracranial hemorrhage), 434 (occlusion of cerebral arteries), 436 (acute but ill-defined cerebrovascular disease), and 437 (other and ill-defined cerebrovascular disease). Patients who had been discharged with a diagnosis of subarachnoid hemorrhage (ICD-9 code 430) or transient cerebral ischemia (ICD-9 code 435) were not targeted. For each surveillance year, we randomly selected a 50% sample for hospital record abstraction. The study was approved by the Institutional Review Board of the University of Minnesota.
Medical Record Abstraction
Using standardized data collection forms and a detailed protocol, trained nurses abstracted a wide range of clinical data from the medical record, including neurological symptoms and signs, diagnostic and therapeutic procedures, and autopsy reports, when available. Ambiguous clinical data were resolved in periodic meetings with study physicians. Neuroimaging reports were photocopied and later abstracted in a standardized fashion by physician reviewers, independent of clinical data.
Stroke Counts and Rates
No method of stroke identification in epidemiological studies, whether it involves the use of all acute cerebrovascular disease discharge codes, a subset of these codes, events meeting certain criteria, or even decisions made by a panel of reviewers, is guaranteed to yield an unbiased representation of time trends. In particular, changes in diagnostic and coding practices over time may undermine attempts to monitor disease trends through hospital discharge codes alone.20 21 To avoid erroneous inference on stroke trends on the basis of a single method, we used five different measures to define acute stroke and computed stroke rates that were based on each of these five measures. Three measures exclusively relied on ICD-9 codes, and two measures incorporated clinical data from the medical records (referred to as diagnostic algorithms or diagnostic criteria).
The three stroke counts that exclusively used ICD-9 codes were (1) cases in which at least one of the target acute cerebrovascular disease codes (431, 432, 434, 436, or 437) was listed among the discharge diagnoses, (2) cases in which any of the aforementioned codes was listed as the first discharge code, and (3) cases in which a code that is more specific for stroke (431, 434, or 436) was listed in any position among the discharge codes.
The two diagnostic algorithms were developed according to concepts described in the National Survey of Stroke22 and in other sources23 24 in an attempt to identify acute stroke consistently through the 1980s.16 Diagnostic algorithms reduce subjective judgment, maintain high levels of specificity (ie, identify definite events), and, most important, attempt to select comparable events from different time periods. The first algorithm, which we call World Health Organization (WHO) criteria, involved the use of the basic WHO definition of acute stroke25 (ie, a new neurological deficit of presumably vascular origin that lasted for ≥24 hours [or until death if the patient died within 24 hours]). The second algorithm, which we call the Minnesota Stroke Survey (MSS) criteria, also required documentation of either one “major” or two “minor” specific neurological deficits. A major deficit was defined as aphasia, two of three body parts (face, arm, or leg) affected unilaterally, visual field deficit, or coma. Minor deficits or signs included dysarthria, apraxia, unsteady gait, one affected body part (face, arm, or leg), or abnormal plantar reflex. Both algorithms excluded events with an overt nonstroke etiology (eg, brain tumor, subdural hematoma) and used autopsy results when available. The autopsy rate, however, did not exceed 6% in any survey year. Events that met the MSS criteria were, by definition, a subset of those that met the WHO criteria.
Events that met the criteria of either algorithm were classified as brain infarction or brain hemorrhage when neuroimaging studies were available or as “undetermined type” in their absence. Brain infarction was further classified as “possibly embolic” if one of the following was documented in the chart: atrial fibrillation, mitral stenosis, intracardiac thrombus, systemic embolus, recent myocardial infarction, or cerebral or cardiac angiography preceding the stroke.
For the past two surveys (1985 and 1990) but not for the 1980 survey, the nurse abstractors searched the index hospitalization record as well as previous records for evidence of a prior stroke. When no evidence of a prior stroke was found, we assumed the event was incident (first ever).
Stroke Mortality in Minneapolis/St Paul
We obtained information on stroke mortality in the survey area for the years 1960 through 1994 from the Minnesota Department of Health. For each year, we identified deaths among Twin Cities residents 30 to 74 years old that were attributed to stroke according the ICD version in effect: for 1960 through 1967, ICD-7 codes 330 through 334; for 1968 through 1978, ICD-8 codes 430 through 438; and for 1979 through 1994, ICD-9 codes 430 through 438.
We adjusted stroke rates (attack rates and incident rates) to the age structure of the US population in 1980 and tested the null hypothesis of no change between the age-adjusted rates of two consecutive survey periods (1985 versus 1980, 1990 versus 1985) using SAS PROC LOGISTIC.26 In addition to the year effect, all models have included age as a covariate. We present here sex-specific, person-based results (ie, counting only one of multiple events per person in a given survey year).
We adjusted the annual rates of stroke mortality to the age structure of the US population in 1980. To reduce year-to-year variation, we computed a 3-year moving average.
Stroke mortality among 30- to 74-year-old residents of metropolitan Minneapolis/St Paul has declined at least since the early 1960s, yet three distinct periods can be recognized for both men and women (Fig 1⇓): a period of slow pace of decline through the early 1970s, a period of rapid decline between the early 1970s and the early 1980s, and a period of slow decline, if at all, since then. None of the transitions from one period to another coincided with a change in the ICD version in effect.
Tables 1⇓ and 2⇓ show for men and women, respectively, the number of hospitalized strokes that were identified by the various counting methods, the population rates, and their 95% confidence intervals. All of the rates, with the exception of those labeled “incident events,” can be considered “attack rates” (ie, the combined rate of first and recurrent events in a given year).
Among men, all measures of age-adjusted stroke rates (including the target ICD-9 hospital discharge codes [431, 432, 434, 436, and 437], only first-listed codes, codes that are more specific for stroke [431, 434, and 436], and events meeting specific criteria) indicate a decline between 1980 and 1985 (Table 1⇑). The decline is estimated to be >20% according to the least-specific measure (all codes) but ranges from 5% to 14% by other measures. Of note, the annual estimates are bound by relatively wide confidence intervals, which are ≈10% to 15% of the rates. For the period of 1985 through 1990, two measures of stroke rate show a decline of 6% to 8%, whereas four other measures, including a measure of stroke incidence, show little change or an increase.
Among women (Table 2⇑), measures that exclusively rely on discharge codes suggest a decline in the rate of stroke between 1980 and 1985 of 4% to 19%. Similar to the data for men, the decline is estimated to be the largest for the least-specific measure (all target codes). In contrast, both WHO and MSS criteria-based rates suggest an increase over that period. For the period of 1985 through 1990, all six measures show some increase of stroke rate ranging from 1% to 10%. Again, the 95% confidence intervals are quite wide, on the order of 15% to 20% of the estimates, and the null hypothesis was not rejected in any comparison of the rate in 1990 with that in 1985.
Trends in age-adjusted stroke incidence between 1985 and 1990, as shown in the last rows of Tables 1 and 2⇑⇑, are further stratified in Figs 2⇓ and 3, respectively, according to the subtype (using a logarithmic scale). There is no evidence of a decline of ischemic stroke or of hemorrhagic stroke in that period. The rate of stroke of undetermined type (in the absence of neuroimaging) has declined, possibly due to greater use of CT and MRI, but the absolute rate of such strokes was low (comparable to the rate of hemorrhagic stroke). The proportion of ischemic strokes (of those identified by the WHO criteria) that were classified as possibly embolic changed from 33% in 1985 to 36% in 1990 among men and from 25% to 40% among women.
Perhaps the most important inference from these results is that any single measure of stroke rates, whether based on discharge codes, a set of criteria, or a classification by a reviewers' panel, might demonstrate trends that are not apparent according to other measures. Although it is impossible to show empirically that any single measure is superior to others with regard to inference on time trends, we comment on the strengths and weaknesses of the measures used here.
The entire list of cerebrovascular disease discharge codes that we targeted may indeed capture most of the events (ie, be most sensitive) but at the cost of including a nontrivial number of hospitalizations for reasons other than stroke. For example, we found that the majority of hospitalizations with a 437 ICD-9 discharge code alone (“other and ill-defined cerebrovascular disease”) are not due to an acute stroke. It should be recalled, however, that official mortality statistics for stroke and trends in stroke mortality that are based on these statistics use the codes that we selected here and a few more. One may argue that trends of stroke mortality in the population, as imprecise as they might be, are best explained by examining coinciding trends in hospital discharge codes because the latter trends, whether real or artifactual, may affect the probability of assigning a cerebrovascular disease code as a cause of death for deceased who had been previously hospitalized. For example, if the rate of acute cerebrovascular disease hospital discharges has declined over time, for whatever reason, it might be expected to observe a parallel decline in cerebrovascular disease diagnoses on death certificates and, consequently, a decline of stroke mortality in the population. Furthermore, the ever-changing practices of assigning diagnoses to hospitalized patients and attributing causes of death to deceased individuals may have common origins, such as contemporary medical knowledge and available diagnostic technology. Whether both of these code-based trends—that of stroke mortality and that of stroke discharges—have an artifactual component is a separate question.
A subset of the codes (eg, excluding ICD-9 code 437) might have the advantage of being more specific but might produce biased trends because the practice of coding, even within the group of acute cerebrovascular disease codes, is not necessarily constant over time. For example, among hospital discharges for acute cerebrovascular disease, we observed a decline in the proportions of ICD-9 codes 436 and 437 and a rise in the proportion of ICD-9 code 434 over the 1980s (data not shown). Counting events in which a target discharge code was listed first might be a better measure of acute stroke than counting any event with a target code, but again, this method offers no remedy to the possibility of time trends in the rank ordering of discharge diagnoses.
Most, if not all, of the code-based methods might be more sensitive (ie, will detect more true strokes in each year) than criteria-based methods such as the diagnostic algorithms used here, but they have inferior specificity. Under certain circumstances, specificity is more important than sensitivity when inferring time trends. For example, if both the availability of information with which events are classified and the case mix of true strokes have not changed over time, then the trend of events that meet specific criteria should parallel the trend for all true strokes, whereas a trend that is based on more sensitive criteria might not. Of course, there is no way of proving that both conditions have been fully satisfied here, and in fact, the quality of medical records (ie, availability of information) has improved over time.
Because there is no empirical method of deciding which measure of stroke rates has yielded the best representation of the (forever unknown) true trend, the best we can do is to search for some consistent patterns. Such an approach assumes, admittedly with no empirical basis, that relative consistency can be equated with validity. Under such an assumption (which is neither better nor worse than the assumption that some specific measure is “the best”), the following inferences can be drawn from our data: for the 1980-through-1985 comparison, most of our measures of hospitalized stroke rate among men suggest a decline on the order of 5% to 10%. The picture among women for that period is unclear, with a sharp contrast between code-based trends showing a decline of similar magnitude to that in men and algorithm-based trends showing the opposite. For both men and women, the target discharge codes, all of which are routinely included in national statistics of stroke mortality, yielded a decline in stroke rate of ≈20% that was not supported in magnitude, and sometimes not supported at all, by the other measures. Those striking declines in the first half of the 1980s might have resulted from artifactual changes in the assignment or coding of discharge diagnoses for cerebrovascular disease. If this was indeed the case, then according to the arguments we presented earlier, a component of the steep mortality decline at that time might also have been artifactual.
Most of our measures of stroke rate (all six in women and four of six in men) refute the hypothesis that the rate of hospitalized stroke declined meaningfully in the survey area between 1985 and 1990. In fact, the pattern for women is consistent with an increase. Stratified analysis according to stroke subtypes showed relatively stable rates of hospitalized ischemic stroke and hemorrhagic stroke in both sexes (Figs 2 and 3⇑⇓). However, because neuroimaging was not used in all cases, not all events could be definitively classified. Within the sample of ischemic strokes, we noted a substantial increase in the proportion of events among women that were classified as possibly embolic. It is uncertain whether this trend reflects a true, alarming change or detection bias.
Broderick et al18 and Brown et al19 proposed a possible explanation of why stroke rates in Olmsted County, Minn, ceased to decline after the late 1970s and why the case-fatality rate of stroke decreased. They suggested that because of the greater use of neuroimaging over time, mild strokes that would not have been recognized previously were detected in more recent periods, artifactually masking a true decline of stroke incidence and artifactually presenting reduced case-fatality rates. This plausible hypothesis was not corroborated by our data. First, we previously reported that improved survival was evident even among stroke patients with substantial neurological deficits that could have been detected by clinicians independent of CT findings.16 Second, the rapid increase in the use of neuroimaging technology in Twin Cities hospitals occurred between 1980 and 1985 (in our sample, from 57% to 75%) with a smaller increase between 1985 and 1990 (from 75% to 82%). According to the aforementioned hypothesis and assuming that indeed the rate of stroke has constantly declined through the 1980s, our data should have underestimated the “true trend” more so for the 1980-through-1985 period than for the 1985-through-1990 period. If at all, we have underestimated more a hypothetical decline between 1985 and 1990 than a hypothetical, comparable decline between 1980 and 1985. Although our findings may contradict, in part, those from Olmsted County for the early 1980s, neither survey strongly corroborates the hypothesis that stroke incidence declined in the latter part of the 1980s.
At least three weaknesses of our methodology should be considered. First, our inference is based on estimation of stroke rates in selected years (1980, 1985, and 1990), and it might be that the rate in one or more of these years does not resemble the rate in adjacent years. Second, although record abstraction followed a detailed written protocol, some inconsistency among abstractors and over time is likely. Third, some stroke victims are never hospitalized,27 yet we had no access to data on nonhospitalized acute stroke in the metropolitan area. The reported trends might be biased as a result of incomplete case determination but only if the rate of nonhospitalized stroke in the survey area among 30 to 74 year olds has changed over time. The magnitude of such bias, if it occurred, is dependent on the magnitude and direction of the difference in the rate of nonhospitalized acute stroke between two survey periods.
The findings reported here on the rates of hospitalized stroke in metropolitan Minneapolis/St Paul and those we reported previously on survival of stroke patients16 are in reasonable harmony with the trend of stroke mortality in the survey area from 1980 through 1990 among residents who were 30 to 74 years old (Fig 1⇑). Much of the mortality decline could be attributed to improved survival of stroke patients. Although a decline in stroke incidence may have occurred in the early 1980s, accounting for some of the steep mortality decline at that time, we found little indication of such a trend in the latter part of that decade—a period of slow to no decline of stroke mortality in the survey area.
This study was supported by grant RO1-HL-23727 from the National Heart, Lung, and Blood Institute. We are in debt to Drs Gregory Burke, Richard Crow, Aaron Folsom, Richard Gillum, Linda Goldman, David Jacobs, and J. Michael Sprafka for their contributions to the design and conduct of this study; to Mary Porter, Elise Brodin, Kristine Bisgard, Tracy Sides, and Albert Tsai for programming assistance; and to the dedicated nurse abstractors who participated in this project. We also thank the hospitals in the Twin Cities for their commitment to this project for more than a decade.
Reprint requests to Eyal Shahar, MD, Division of Epidemiology, School of Public Health, University of Minnesota, 1300 S Second St, Suite 300, Minneapolis, MN 55454-1015. E-mail email@example.com.
- Received August 21, 1996.
- Revision received October 25, 1996.
- Accepted November 12, 1996.
- Copyright © 1997 by American Heart Association
Klag MJ, Whelton PK, Seidler AJ. Decline in US stroke mortality: demographic trends and antihypertensive treatment. Stroke. 1989;20:14-21.
Lanska DJ, Peterson PM. Geographic variation in the decline of stroke mortality in the United States. Stroke. 1995;26:1159-1165.
Bonita R, Stewart A, Beaglehole R. International trends in stroke mortality: 1970–1985. 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.
Garraway WM, Whisnant JP, Drury I. The changing pattern of survival following stroke. Stroke. 1983;14:699-703.
Ahmed OI, Orchard TJ, Sharma R, Mitchell H, Talbot E. Declining mortality from stroke in Allegheny County, Pennsylvania: trends in case fatality and severity of disease, 1971-1980. Stroke. 1988;19:181-184.
Howard G, Toole JF, Becker C, Lefkowitz DS, Truscott BL, Rose L, Evans GW. Changes in survival following stroke in five North Carolina counties observed during two different periods. Stroke. 1989;20:345-350.
Modan B, Wagener DK. Some epidemiological aspects of stroke: mortality/morbidity trends, age, sex, race, socioeconomic status. Stroke. 1992;23:1230-1236.
McGovern PG, Pankow JS, Burke GL, Shahar E, Sprafka JM, Folsom AR, Blackburn H. Trends in survival of hospitalized stroke patients between 1970 and 1985: the Minnesota Heart Survey. Stroke. 1993;24:1640-1648.
May DS, Casper ML, Croft JB, Giles WH. Trends in survival after stroke among Medicare beneficiaries. Stroke. 1994;25:1617-1622.
Shahar E, McGovern PG, Sprafka JM, Pankow JS, Doliszny KM, Luepker RV, Blackburn H. Improved survival of stroke patients during the 1980s: the Minnesota Stroke Survey. Stroke. 1995;26:1-6.
Broderick JP, Phillips SJ, Whisnant JP, O'Fallon WM, Bergstralh EJ. Incidence rates of stroke in the eighties: the end of the decline in stroke? Stroke. 1989;20:577-582.
Brown RD Jr, Whisnant JP, Sicks JD, O'Fallon WM, Wiebers DO. Stroke incidence, prevalence, and survival: secular trends in Rochester, Minnesota, through 1989. Stroke. 1996;27:373-380.
Walker AE, Robins M, Weinfeld FD. The National Survey of Stroke: clinical findings. Stroke. 1981;12(suppl I):I-32-I-44.
National Heart, Lung, and Blood Institute, Atherosclerosis Risk in Communities (ARIC) Study Protocol, Manual 3: Surveillance Component Procedures. Chapel Hill, NC: ARIC Study Coordinating Center, School of Public Health, University of North Carolina; 1991.
SAS/STAT User's Guide, Version 6. Cary, NC: SAS Institute; 1989.
Feng E-J, Orencia AJ, Whisnant JP. Outcome and pathogenesis of stroke is conditioned by place of stroke occurrence. J Stroke Cerebrovasc Dis. 1993;3:3-8.