Patients with acute stroke often arrive at the hospital too late to receive the maximum benefit from these new stroke therapies. Efforts to reduce delay time of therapy for acute stroke may be more effective if the factors that delay hospital arrival are identified and targeted for specific intervention; however, only a few studies have extensively examined this issue in the United States [6-10]. In contrast, factors influencing hospital arrival time for acute myocardial infarction have been studied in detail [11-21]. Several factors that delay hospital arrival after acute stroke parallel those for acute myocardial infarction, including living alone [8, 12], being at home during symptom onset [7, 11, 12], and first contacting a physician rather than an emergency service [7-9, 12-17]. By using medical record review, we investigated factors that influence the time from stroke onset to patient arrival at metropolitan Minneapolis-St. Paul hospitals.

Methods

Top Methods Results Discussion Author & Article Info References

Sample Selection

The Minnesota Stroke Survey is a population-based surveillance study of acute cerebrovascular disease [22]. The target population comprises residents 30 to 79 years of age living in the seven-county metropolitan area of Minneapolis-St. Paul, Minnesota. We report on factors that influenced hospital arrival time during the early 1990s by using combined data from surveys from three calendar years (1991, 1992, and 1993). Discharge diagnosis lists based on the International Classification of Diseases, 9th Revision (ICD-9), stroke codes were obtained from 23 of 25 acute care hospitals in the Minneapolis-St. Paul area for each surveillance year. The two hospitals that did not provide data have low patient volume (an estimated two to four valid patients with stroke per hospital per year). The sampling frame included all Minneapolis-St. Paul residents 30 to 79 years of age who were discharged with an acute cerebrovascular disease code (ICD-9 codes 430, 431, 432, 434, 436, and 437). For each surveillance year, a 50% random sample was selected from the listings for detailed record abstraction. No data were available for patients who were not hospitalized. Previous studies have estimated that 81% of all patients with a first stroke event were hospitalized, and patients with fatal strokes were less likely to be hospitalized than those with nonfatal strokes (76% and 89%, respectively) [23].

Data Collection

Trained nurses used standardized data collection forms to abstract clinical data from the medical record, including date and time of symptom onset, data and time of admission to the hospital, demographic characteristics, medical history, symptoms at stroke onset, physician examination findings, stroke severity, and stroke type. Neuroimaging and autopsy reports were photocopied and abstracted by study physicians who did not have knowledge of the patient's other medical record data. The study was approved by the institutional review board of the University of Minnesota.

Stroke Validation and Classification

Because factors influencing delay time were assumed to be unrelated to stroke type, we included all stroke types in the sample. Events were included when the patient had a new neurologic deficit that lasted at least 24 hours or until death. Patients with brain tumors, subdural hematomas, and subarachnoid hemorrhages were excluded. Autopsy results were available for only 9% of patients who died while they were hospitalized. Accordingly, autopsy results were used to confirm the diagnosis of stroke for less than 1% of patients (n = 16). Stroke events were classified as "hemorrhagic" or "ischemic" on the basis of neuroimaging results and autopsy reports or as "undetermined" when neither data source was available. Ischemic stroke was divided into "possibly embolic" and "nonembolic" subtypes. A stroke was considered possibly embolic if ICD-9 code 434.1 (cerebral embolism) was listed among the discharge diagnoses, if the medical record indicated that the stroke was definitely or probably embolic (including embolism from the carotid arteries), or if one or more of the following conditions was documented in the medical record: atrial fibrillation, mitral stenosis, intracardiac thrombus, systemic embolus, recent myocardial infarction, or cerebral or cardiac angiography that closely preceded the stroke.

Patients were excluded if symptoms developed while the patient was hospitalized for another condition (n = 294) or if the patient had been transferred from another hospital (n = 41). During the initial stages of the medical record abstraction, we did not collect information on whether the patient was admitted to the hospital through the emergency department; these patients were also excluded (n = 216). At the time of analysis, data collection was 77% complete. The final available sample size was 1895 hospitalizations for acute stroke.

Variables

The main variable of interest, delay time, was defined as the time from the onset of new neurologic symptoms to the time of hospital arrival. Time of hospital arrival was defined as the time the patient presented to the emergency department or the hospital (if the patient was not admitted through the emergency department). The time of symptom onset was recorded in the medical record of 70% of the patients (n = 1334). The delay time was calculated by subtracting the recorded time of symptom onset from the hospital arrival time when both times were available. For the remaining patients (n = 561), the time of symptom onset was not recorded in the medical record; therefore, we approximated the delay time on the basis of all available information (<24 hours, 24 hours to <48 hours, 48 hours to <72 hours, and 72 hours to <6 weeks).

Demographic characteristics included age, sex, and ethnicity. Ethnicity was categorized as white, African American, Asian/Pacific Islander, and other. We also identified whether the patient was enrolled in a health maintenance organization, whether the patient lived alone or in a nursing home before the stroke, whether do-not-resuscitate or do-not-intubate orders existed before hospital admission, and whether the patient was admitted to the hospital through the emergency department. A history of the following conditions was recorded: transient ischemic attack, stroke, myocardial infarction, diabetes, hypertension, hypercholesterolemia, whether the patient ever smoked, and daily use of aspirin or warfarin. The level of comorbidity was estimated by using a modified Charlson comorbidity index, excluding two conditions specific to acute stroke (cerebrovascular disease and hemiplegia) [24]. The patient's prestroke functional status was measured by determining whether the patient was dependent in at least one activity of daily living (bathing, continence, feeding, going to the toilet, and transfer into or out of a bed or chair) before the onset of the stroke [25].

The presence of the following symptoms at stroke onset was recorded: limb or facial weakness, impaired speech, impaired vision, dizziness, unsteadiness, tingling, seizures, syncope, headache, and vomiting. The time of symptom onset was classified by the quarter of the day that the stroke occurred (morning = 6 a.m. to noon, afternoon = noon to 6 p.m., evening = 6 p.m. to midnight, night = midnight to 6 a.m.). The date of symptom onset was used to characterize whether the stroke occurred on a weekend (Saturday or Sunday).

Several variables were used to indicate the severity of the stroke at presentation. The level of consciousness was assigned one of three categories (unconscious, conscious but abnormal, or normal). The level of disability was estimated by using the Rankin score (from 1 = no disability to 5 = severe disability) [26]. The following findings on physician examination were also noted: weakness or paralysis of the limbs and face, visual defects, ataxia, sensory loss, aphasia, dysarthria, impaired swallowing, abnormal findings on pupil examination, and abnormal plantar (Babinski) reflex.

Statistical Analysis

Parametric survival regression analysis based on an accelerated failure time model was used to examine the influence of the explanatory variables on delay time. Patients with a calculated delay time were treated as uncensored data. Patients with an approximated delay time that indicated arrival within 24 hours of symptom onset were treated as left-censored at 24 hours. Patients with an approximated delay time that indicated arrival within 24 to 48 hours, 48 to 72 hours, and 72 hours to 6 weeks were treated as interval-censored data. The generalized model and three models nested within the generalized model (exponential, Weibull, and log-normal models) were compared by using the likelihood ratio statistic. The two best-fitting models were the generalized model and the log-normal model. The conclusions did not differ between the two models, and the results from the log-normal model are presented. The ß coefficients estimated from this model were converted to the percentage difference in delay time associated with a one-unit increment in the explanatory variable X [% Delta Delay/Delta X = (e^ß- 1) x 100]. For continuous variables, this value indicates the percentage difference in delay time associated with a one-unit increment in the explanatory variable. For indicator variables, this value indicates the percentage difference in delay time in the presence of the indicator. The data were analyzed by using SAS PROC LIFEREG [27]. Because the delay times for patients may have been clustered within hospitals, we also used a mixed-effects model (SAS PROC MIXED) to examine clustering by hospital for the 70% of cases with calculated delay times (data not shown). We found no evidence of within-hospital correlations (intraclass correlation coefficient, 0.000), and SEs were almost identical to the results obtained with the fixed-effects model.

Results

Top Methods Results Discussion Author & Article Info References

Patient Characteristics

The characteristics of the study patients are shown in Table 1. The mean age was 66 years, and almost half of the sample was female. Eighty-eight percent of the patients were white, 7% were African American, and 2% were Asian/Pacific Islander. One quarter of the patients lived at home alone, and almost half were enrolled in a health maintenance organization. Most patients had a history of another important medical condition. Almost half of patients had a history of transient ischemic attack or stroke, and one quarter had a history of diabetes. One quarter of patients were receiving a daily dose of aspirin before admission; 1 in 10 was receiving warfarin. The mean value of the Charlson comorbidity index was 1.7 (range, 0 to 12), indicating that many patients had at least one additional medical condition that was important enough to influence survival during hospitalization. One quarter of patients were dependent in at least one activity of daily living before symptom onset, and 1 in 20 had do-not-resuscitate or do-not-intubate orders before admission to the hospital.

The timing of symptom onset was not evenly distributed. Fewer strokes occurred at night, and most strokes occurred in the morning. Most patients (81%) were admitted to the hospital through the emergency department. Limb or facial weakness was the most prevalent symptom at stroke onset (69%), although most patients also had speech impairment (56%). One third of patients developed unsteadiness, and one quarter reported tingling sensations. Seizures and syncope were relatively rare (<4% of patients). Headache and vomiting were present in one fifth and one tenth of patients, respectively.

The mean Rankin score was 3.4 ± 1.0, indicating that most patients had substantial disability on presentation. Eleven percent of patients were unconscious, and 42% were conscious but had some degree of abnormal mentation. Arm weakness or paralysis was the most common neurologic deficit identified on physician examination; this condition was present in two thirds of patients. Less than half of patients had dysarthria, and approximately one third had aphasia, sensory loss, and ataxia. One fifth of patients had visual field defects, and one quarter had an abnormal result on pupil examination. Almost half of all patients had an abnormal plantar (Babinski) reflex. Finally, 11% of patients were classified as having had hemorrhagic stroke, 85% as having had ischemic stroke, and 4% as having had a stroke of undetermined type. One third of strokes were possibly embolic, and more than half were nonembolic.

Delay Time to Hospital Arrival

Figure 1 shows the delay time for all patients in broad categories. Patients with approximated delay times tended to have longer delays, and less than 40% of patients arrived within 24 hours of symptom onset. In contrast, 90% of patients with calculated delay times arrived within 24 hours of symptom onset. For the 70% of patients with calculated delay times, Figure 2 gives more precise categories of delay times. Approximately half of these patients arrived within 3 hours of symptom onset, and two thirds arrived within 6 hours.

View larger version (11K): [in this window] [in a new window]   Figure 1. Frequency distribution of delay time for patients hospitalized for acute stroke, with calculated delay times (n = 1334) and approximated delay times (n = 561).

View larger version (7K): [in this window] [in a new window]   Figure 2. Frequency distribution of delay time for hospitalized patients with acute stroke and calculated delay times only (n = 1334).

Patient Characteristics Predicting Delay Time

Table 2 shows the patient characteristics that independently predicted delay time. Asian/Pacific Islander patients had a 107% longer delay time than white patients. Patients with dependence in any activity of daily living before the stroke had a 36% longer delay time compared with patients who were independent in all five activities of daily living. Three symptoms (impaired vision, unsteadiness, and headache) were independently associated with longer delay. Finally, patients who did not have the time or day of symptom onset recorded in the medical record had substantially longer delay times. For example, patients for whom the time of symptom onset was not recorded in the medical record had a 383% longer delay than patients for whom the time of symptom onset was recorded. Three other characteristics reached borderline statistical significance: history of diabetes, preexisting do-not-resuscitate or do-not-intubate order, and ataxia at presentation.

Several patient characteristics were associated with shorter delay. Delay time was substantially shorter with admission through the emergency department (46%). Other characteristics associated with shorter delay included syncope or seizures at stroke onset; possibly embolic, hemorrhagic, or undetermined stroke type; and a history of myocardial infarction. Increasing disability at presentation (measured by the Rankin score) was linked to shorter delay. Patients who were conscious but had abnormal mental status had a 16% shorter delay than patients whose mental status was normal. Unconsciousness at presentation, female sex, tingling at stroke onset, and visual defect and dysarthria as determined by physician examination were of borderline statistical significance.

Eighteen observations were identified that potentially influenced the fit of the model. The conclusions were similar after the analysis was done without these observations. The single discrepancy was that higher levels of the comorbidity index were associated with longer delay. For each unit increment in the comorbidity index, the model predicted a 4% longer delay time (95% CI, 0% to 8%).

Discussion

Top Methods Results Discussion Author & Article Info References

In our study, most patients with acute stroke arrived at the hospital too late to receive the maximum benefit from new stroke therapies. Most patients arrived at the hospital too long after symptom onset to be eligible for tissue plasminogen activator therapy, and many of those who did arrive in time would have been ineligible because of the characteristics of their stroke or medical history. A substantial proportion of patients arrived within 3 to 6 hours after symptom onset, and these patients may benefit from emerging therapies, such as neuroprotective agents. However, many patients still arrive more than 12 hours after the onset of symptoms, and a substantial number arrive days later. Reductions in delay time will be needed if the maximum number of patients are to receive the full benefit of new stroke therapies.

Effective public health interventions to reduce delay time should be based on an understanding of the factors that delay hospital arrival for acute stroke patients. In our study, ethnicity was an important patient characteristic that predicted delayed hospital arrival. In particular, patients of Asian/Pacific Islander descent had substantially longer delay than white patients, although the absolute number of Asian/Pacific Islander patients in this sample was small. The lack of a statistically significant difference for patients from other ethnic groups may also reflect the relatively small numbers of these patients. One study of acute stroke did not find an association with ethnicity [7], but longer delays have been shown for African-American and Hispanic patients with acute myocardial infarction [13, 16]. Longer delay may arise from cultural factors or the lack of a usual care provider [28, 29]. Patient age and sex were also not significant factors influencing delay time in this analysis, although female sex was associated with shorter delay and was of borderline statistical significance. Although one study of stroke suggested that increasing age might be associated with longer delay [8], that study did not control for potentially confounding variables, such as previous functional dependence among older patients. In contrast, both older age and female sex have been associated with longer delay for presentation with myocardial infarction [11, 12, 15, 19, 30].

Members of health maintenance organizations were not less likely to delay hospital arrival. Some managed care settings require patients to obtain authorization before visiting the emergency department, and these regulations have been seen as a potential barrier to rapid emergency care for patients with the symptoms of acute and potentially life-threatening disease [20]. However, we found no evidence that managed care contributed to increased delay for patients with acute symptoms of stroke.

Patients with preexisting health concerns (such as dependence in activities of daily living or comorbid conditions) also had longer delay times. Stroke can present with a bewildering array of symptoms, and new symptoms of acute stroke might be mistakenly interpreted by some patients as related to preexisting health problems [18, 31]. Similar results were reported for acute myocardial infarction, in which a history of cardiac risk factors (such as diabetes) lengthened delay [30]. It is also possible that dependence in activities of daily living (such as feeding or transfer) interferes with the ability to seek care for new symptoms [18].

Milder and less visible symptoms may also lengthen the delay in seeking medical attention [18]. Symptom severity is important for patients with myocardial infarction, and shorter delays have been described for patients with more severe symptoms [17]. Patients with milder strokes had substantially longer delay, as assessed by the Rankin disability score at presentation. Various symptoms at stroke onset also delayed arrival at the hospital, including impaired vision, unsteadiness, and headache. Syncope and seizures were associated with shorter delay. Unsteadiness and headache occur in a wide array of nonspecific and non-life-threatening medical conditions. Patients may consider these symptoms less serious and delay longer as a consequence. In contrast to these results, an earlier study suggested that headache and vomiting shortened delay, probably because these symptoms are associated with the onset of hemorrhagic stroke [9]. However, we controlled for stroke type and for several measures of stroke severity.

Familiarity with stroke symptoms and the consequences of stroke may also influence delay time [18], but recent evidence suggests that knowledge alone is not enough to change care-seeking behavior [10]. We found no evidence of shorter delay for patients with previous transient ischemic attack or stroke or daily use of aspirin or warfarin. A similar result has been found for acute myocardial infarction, in which a history of coronary heart disease did not reduce delay [15, 20, 30]. Of interest, previous myocardial infarction was associated with shorter delay for these patients with stroke, but previous stroke was not.

These data are consistent with research showing delayed hospital arrival for infarction compared with hemorrhagic stroke [6, 9]. In our analysis, hemorrhagic and possibly embolic strokes were associated with shorter delays. The more dramatic and abrupt onset of hemorrhagic and embolic strokes may explain this finding [6, 9]. Another finding similar to that of previous studies of stroke and myocardial infarction was that nonemergency admission delayed arrival after the onset of acute stroke [7, 9, 12, 14, 17, 20]. We did not find evidence that living alone, nocturnal onset, or weekend onset influenced delay, although these factors have been shown to increase stroke delay time in other studies [7-9]. The large number of variables analyzed may explain this apparent discrepancy. For example, chronobiological variation in acute stroke may be associated with variables (such as stroke type) for which we controlled in this analysis [32].

Several limitations of our study deserve mention. First, although a time of symptom onset was available in the medical record for 70% of patients, this recorded time is susceptible to measurement error by both patients and providers. For the remaining 30% of patients, only an approximate time was available. Although the inclusion of all patients in the regression model was a major strength of our analysis, the lack of a precisely measured time of symptom onset merits a cautionary note. Second, no information was available about whether the patient was at home or at another location when the stroke occurred. In both myocardial infarction and stroke, symptom onset at home has been associated with delayed hospital arrival [7, 12]. Finally, these data are more than 4 years old, and patterns of delay time may have evolved. However, no widespread public education efforts to reduce delay time for acute stroke have been made in recent years, and no evidence exists that delay times have changed greatly.

Many patients with stroke arrive at the hospital too late to receive the maximum benefit from medical care. Although only a few patients with acute stroke will probably be eligible for thrombolytic therapy, many other advances in diagnosis and treatment have improved care for stroke [33, 34]. Early neurologic attention has been related to improved functional outcome and shorter hospitalization [35]. Rapid emergency care and admission to a hospital unit that specializes in cerebrovascular disease is currently recommended for all patients with acute stroke [33, 34]. Efforts to reduce delays in hospital arrival can maximize the effectiveness of therapy by specifically targeting persons at risk for longer delay. As a result, more patients should become eligible for new stroke therapies that require early administration or for enrollment in clinical trials of the benefits of emerging stroke therapies.

From the School of Public Health, University of Minnesota, Minneapolis, Minnesota.

Drs. Doliszny, Shahar, McGovern, Arnett, and Luepker: Division of Epidemiology, School of Public Health, University of Minnesota, 1300 South Second Street, Suite 300, Minneapolis, MN 55454-1015.