Population-Based Intervention to Reduce Prehospital Delays in Patients With Cerebrovascular Events FREE

According to the World Health Organization, approximately 15 million people worldwide experience a stroke per year.¹ Of these, 5 million die and another 5 million are left permanently disabled, placing a substantial burden on family and community. The severity of stroke-related disability can be reduced if timely and appropriate treatment is received.^2- 3 Patients with ischemic stroke may be eligible for treatment with intravenous thrombolytic therapy such as recombinant tissue plasminogen activator, which should be administered within 3 hours of symptom onset.² Thrombolytic therapy potentially reduces disability from stroke by approximately one-third.^4- 5 However, the estimated percentage of patients with stroke receiving thrombolysis worldwide is tiny (<1%), with only some regions achieving slightly higher percentages (3%-4%).^6- 7

One of the major barriers to the application of thrombolytic therapy is prehospital delays between symptom onset and arrival at the emergency department.^2,8 Data from the Paul Coverdell National Acute Stroke Registry in the United States showed that of the more than 17 600 patients, 48% of those with known time of onset (45%) arrived within 2 hours.³ Of all patients, an estimated 25% arrive within 3 hours of symptom onset.⁶ Despite decades of research and public education campaigns aimed at decreasing patient delay times, most patients still do not seek treatment in a timely manner.⁸ Predictors for prehospital delays include symptoms considered urgent or not, perceived awareness of having a stroke, initial call to the emergency medical services (EMS), stroke severity, ethnicity, and type of stroke.^6,9- 12 Among these, potentially modifiable predictors are recognizing stroke symptoms and calling the EMS. Only approximately a quarter of patients are actually aware of having a stroke.¹⁰ If patients called the EMS immediately, the expected rate of thrombolytic therapy would increase to an estimated 29%.⁶

Many health care systems and nongovernment agencies are currently trying to reduce prehospital delays for stroke. However, in order not to spend resources on ineffective interventions—no matter how attractive their face validity—there is an urgent need for scientific evidence. Interventions to reduce prehospital delays in patients with stroke have focused on education campaigns for the public, training programs for paramedical staff, or helicopter transfers to the hospital.⁹ Although most studies showed some effect of the respective intervention(s), a number of methodological issues such as lack of a control group or a nonrandomized design severely limited the validity of the results. In addition, spillover effects need to be taken into account: for example, a person receiving the intervention may talk to friends or colleagues who did not receive the information, compounding its effect. The use of a cluster-randomized study design can compensate for this effect. In addition, Hickenbottom et al¹³ showed that solely increasing patient knowledge about warning signs of stroke is not effective in decreasing delay times. The objective of the present study was, therefore, to evaluate the effectiveness of an education- and reminder-based intervention to reduce prehospital delays, using a cluster-randomized design based on zip code areas in a large urban metropolis.

The Berlin Acute Stroke Study (BASS) was a cluster-randomized controlled trial with zip code areas as clusters randomized to intervention and control groups.^12,14 The 48 zip codes constituted the catchment area for the 3 participating hospitals. Of the 3 hospitals, 2 were university hospitals and 1 was a teaching hospital with a stroke unit. The city of Berlin, Germany, has a total of 12 districts, with the study zip code areas located in 9 of them (Table 1).¹⁵ The primary end point of the study was prehospital time, which was assessed in all patients admitted with stroke or transient ischemic attack (TIA) following the intervention phase (Figure 1). The study was approved by the ethics committee of the Charité University Medical Center, Berlin.

The intervention was performed on the individual level. It consisted of a personalized educational letter describing warning signs and symptoms of stroke, as well as appropriate help-seeking behaviors.¹⁴ Key elements were a bookmark and sticker with main stroke symptoms and the telephone number of the EMS. The sticker served as reminder and could be placed, for example, on the telephone or other prominent places in the home. In case of a stressful emergency, we considered it imperative that the relevant information was easily accessible for patients and proxies. The design of the intervention program was based on the results of our previous cross-sectional study.¹² We identified variables associated with prehospital delays in patients with stroke. Whereas some variables such as age or severity of stroke symptoms were nonmodifiable, we considered others to be a target for an intervention program. Particularly, calling the EMS as well as considering symptoms as urgent were judged to be potentially modifiable by our study group. The health education program thus focused on providing information on the emergency telephone number and describing stroke symptoms warranting immediate action by patients. All inhabitants aged at least 50 years in the zip code areas of the intervention group received the health education material by mail. We used the setting of our study to additionally determine risk factor knowledge and prevalence of stroke in the population. Inhabitants of the intervention clusters, therefore, received a standardized questionnaire on stroke risk factors and symptoms.¹⁴ No intervention was performed in the control areas.

We chose a cluster-randomized design to account for a “spillover” effect between intervention and control groups. In addition, other factors such as socioeconomic background tend to be correlated in the same cluster, which is compensated for in this type of design. The zip code areas were randomly allocated to the 2 groups using a computer-generated randomization list. Allocation was based on zip code clusters rather than individuals. The sequence was concealed until the intervention was assigned. Blinding was not feasible for those administering the intervention.

We chose the age of 50 years as the cutoff point for inclusion because the risk of stroke and TIA was considered too low for those younger than 50 years.¹⁶ The Berlin vital statistic office provided names and addresses of inhabitants in the zip code areas of the intervention group. A total of 75 720 households in these areas met the age criteria and received the intervention. Outcomes were assessed in all patients with stroke or TIA admitted to 1 of the 3 participating hospitals. Trained assessors documented socioeconomic factors such as age and sex, risk factors, medical history, and time of symptom onset from the medical records. In patients without exact time of onset in the medical records, we estimated time of onset.¹⁷ When symptom onset was documented as “morning,” “midday,” “afternoon,” “evening,” or “night,” we assumed time of onset as 9 AM, 12 PM, 3 PM, 9 PM, or 3 AM, respectively. When symptom onset was described as “while sleeping,” we assumed midnight (12 AM) of the respective night as time of onset, as well as when “day of onset” but not the time was documented in the medical records. No documentation of day was designated as missing data. The assessors were blinded to group assignment and unaware of the treatment allocation.

The primary outcome was prehospital time on the individual level. Predefined secondary outcomes were the proportion of patients receiving thrombolysis and the percentage of deaths during the hospital stay.

Sample size was determined for the primary outcome prehospital time. We assumed a reduction in geometric mean (equivalent to the arithmetic mean of the log of the prehospital time) of 40% by the intervention. To account for the effect due to the clustered structure of the data, a simulation was performed with bootstrap samples from the data set of our previous study,¹² which contained 144 zip code clusters with varying cluster sizes (mean, 3 patients each). The simulation model was performed with 1000 repetitions, resulting in a number of 400 patients required in each group to demonstrate a 40% reduction in mean prehospital time with a power of 85.3% (α = .05, 2-sided).

For prehospital time interval, we fitted a log-normal survival regression model with frailty terms shared by inhabitants of the same zip code area. A shared frailty model is a random effects model for time-to-event data, where the random effect (the so-called frailty, a purely statistical term in this context) represents a cluster-specific constant that in our case is best understood as deviation of the mean prehospital time in the specific zip code area from the overall mean of all study areas without intervention. We preferred parametric models to the more widespread semi-parametric Cox proportional hazards models because they allow a direct calculation of delay or acceleration of failure times and present the corresponding acceleration factors with 95% confidence intervals (CIs). For example, an acceleration factor of 0.73 corresponds to a 27% reduction of delay. Among the parametric accelerated failure time models, the deviance (−2 log likelihood) indicated a substantially better fit of the log-normal model compared with any other alternative model of the exponential, Weibull, or log-logistic model. The log-normal model was easy to interpret because it assumed that the apparent skewness of the delay time distribution can be transformed to a normal distribution by taking the logarithm, which is easily visualized. In the log-normal model, we compared the 2 random groups adjusting for age, sex, overweight/obesity, and smoking. We added potential interaction terms to the model one by one and report the resulting final model, including the interaction term sex × intervention. For the secondary outcomes thrombolysis and death, we used generalized estimating equations to calculate adjusted proportions with 95% CIs and to compare these proportions between both groups. All tests for significance were 2-sided; the significance level was α = .05. Statistical analyses were performed using SPSS, version 11.0 for windows (SPSS Inc, Chicago, Illinois), and SAS versions 8.1 and 9.1.3 (SAS Institute Inc, Cary, North Carolina), as well as STATA 10.0 (StataCorp, College Station, Texas) for accelerated failure time models.

Between February and April 2004, we sent the educational material by post to 75 720 inhabitants in the 24 intervention cluster areas (Figure 1). No intervention occurred in the control areas. Between April 2004 and December 2005, a total of 2316 patients were admitted to 1 of the 3 hospitals for stroke or TIA. Of these patients, 741 lived in the control zip code areas and 647 in the intervention areas, 928 lived in other zip code areas and were excluded from the analyses. Table 1 gives the characteristics of zip code areas and patients from the intervention and control groups. Socioeconomic factors, risk factors, and medical history of patients were similar between the 2 groups.

Median prehospital time was 8 hours 34 minutes in the control group compared with 6 hours 39 minutes in the intervention group (Table 2). A prehospital time interval of 2 hours or less was attained by 22% of patients in the control group compared with 26% in the intervention group. Table 2 gives the percentages of patients in both groups arriving 3 and 4 hours or less, respectively. Figure 2 depicts the distribution of prehospital time intervals, stratified by sex.

In the log-normal model with shared frailty and prehospital time as dependent variable, there was a significant interaction between sex and intervention (P = .02) (Table 3). The intervention significantly reduced prehospital time in women by about 27% (acceleration factor, 0.73; 95% CI, 0.58-0.94). There was no significant effect of the intervention in men. Also, women in the control group revealed a trend to arrive later in hospital compared with men in the control group (acceleration factor, 1.22; 95% CI, 0.97-1.55). Smokers had a significantly increased prehospital delay compared with nonsmokers. Obesity remained in the final model but was not associated with prehospital time.

With regard to secondary outcomes, an adjusted 2.3% (95% CI, 1.4%-3.7%) of patients in the control group received thrombolytic therapy compared with 2.9% (95% CI, 1.5%-5.7%) in the intervention group (P = .31). During hospital stay, 3.2% (95% CI, 1.7%-5.8%) died in the control group compared with 2.6% (95% CI, 1.4%-4.8%) in the intervention group (P = .49).

Complete information about symptom onset was available for 30.6% (n = 227) in the control and 30.5% (n = 197) in the intervention group. Time periods such as “morning,” “midday,” “afternoon,” “evening,” or “night” were available for 27.7% (n = 205) in the control and 29.5% (n = 191) in the intervention group, and “while sleeping” or “day of onset” were documented in 28.6% (n = 212) and 28.1% (n = 182) of patients, respectively. No information on time of onset was available for 13.1% (n = 97) of patients in the control and 11.9% (n = 77) in the intervention group. There was no significant difference in missing values between the groups or between men and women. In both the control and intervention group, 14 patients had to be excluded because only time of onset but not time of emergency department arrival was available.

This population-based intervention effectively reduced prehospital delays in women with stroke or TIA by approximately 27% but showed no such effect in men. To our knowledge, this is the first study using a cluster-randomized design showing a reduction in prehospital delays in patients with cerebrovascular disease.

In a systematic review (2004), Kwan et al⁹ summarized studies of interventions aimed at reducing prehospital and intrahospital delays in patients with stroke. They found 10 nonrandomized studies: 4 before-and-after studies, 5 observational studies, and 1 nonrandomized clinical study. Most studies found a significant reduction in prehospital delays. However, owing to methodological issues, bias and confounding cannot be excluded and causal inference cannot be assumed. For example, the effect of secular trends on any observed effect is difficult to assess without control group. In the before-and-after study by Morgenstern et al,¹⁸ a parallel group was included, allowing the comparison of 2 communities. Delay time decreased in both communities with no significant difference between groups. Most interventions used a multifaceted approach, for example, combining mass and small-size media as well as training paramedics and physicians. Albeit multifaceted interventions are often the most effective in behavioral change, this renders the evaluation of single program components difficult.

In patients with myocardial infarction, efforts to reduce prehospital and intrahospital delays have been undertaken for almost 2 decades. Caldwell et al¹⁹ performed a systematic review focusing on the effect of mass media campaigns in patients with acute myocardial infarction for the years 1985 to 2000. They found 8 intervention studies meeting inclusion criteria, with 3 reporting successful interventions. High-risk populations and those with confirmed myocardial infarction responded more quickly. However, only 2 studies used control groups and randomized designs, and neither showed success in reducing prehospital delay.^20- 21 In the Atherosclerotic Risk in Communities Study (ARIC), no statistically significant reduction was observed in the proportion of patients with delays of 4 hours or more between 1987 and 2000.²²

We found a significant effect of the intervention in women but not in men. Most studies aiming at reducing prehospital delays have neither reported results for men and women separately^9,19 nor analyzed the interaction between intervention and sex. In most studies in patients with stroke, female sex has not been associated with increased prehospital delays.^8,12,23 In patients with acute coronary syndromes, increased prehospital delays in women have been found in some studies.^8,13 The most important sex-specific factor for delays appears to be that women do not want to trouble anyone.^23- 24 Our intervention might thus have “allowed” women to seek help in case of an acute event. Another explanation for the observed effect of the intervention may be that women are more susceptible to health-related recommendations.²⁵ However, other research methods such as qualitative assessments may be needed to explain the difference in response to the intervention between men and women. In addition, the positive finding of a reduction in prehospital delays in women is based on a subgroup analysis. Subgroup analyses have inherent limitations such as the lack of power and have to be interpreted with caution.²⁶ In particular, it has to be kept in mind that sex-specific intervention effects were not anticipated when the trial was launched and thus are hypothesis generating rather than confirmatory. An independent validation is required before further conclusions can be drawn.

Another factor in the effectiveness of our intervention compared with other parallel-group studies might have been the use of an attenuated high-risk vs an entirely population-based approach. Moser et al⁸ stated that no published interventions have targeted high-risk patients with the goal of reducing prehospital delays in stroke to date. Inhabitants in our study had to be at least 50 years of age to receive the intervention. Although we did not use risk stratification methods including the assessment of risk factors, increasing age is per se a major risk factor for stroke and easy to determine. Perceived risk of cardiovascular events in oneself or proxies will certainly increase susceptibility to interventions.²⁷ In addition, we included a personalized letter from the major university hospital in the city. Often patients feel that they need authorization to call the EMS,²⁸ which our letter might have provided. Finally, we used a sticker as reminder, since patients or their proxies may be too stressed in the case of an immediate emergency to recall both stroke symptoms and the telephone number of the EMS. Studies have shown that knowledge of emergency telephone numbers decreases with age, and in age groups older than 60 years, approximately 25% do not know the emergency number to call.²⁹

There are a number of limitations to the present study. Sex may be a surrogate for other factors like living alone or education that could not be determined in this study. We cannot exclude that some patients were admitted to other hospitals, and we have no information on these patients. In addition, we did not have information on patients dying prior to reaching the hospital because only patients admitted to one of our participating hospitals could be included in the analysis. Also, patients reporting symptoms on waking up may have had their symptom onset hours beforehand.¹⁷ Thus, left censoring may have introduced bias in either direction, although we assume that it mainly reduced the power of the study and the bias was small. Another limitation is that the time of event was completely missing in 12% to 13% of patients. However, there were no differential missing values between the 2 groups. Other studies report similar rates of missing values.^18,21 Missing data may be due to lack of questioning or recording by the emergency personnel, the patient's condition, or patient uncertainty about time of symptom onset. Because of the limited precision of time assessments, delay times in our study may be misleading. Although the percentage of patients arriving within 3 hours is in accordance with other studies, mean and median delay times may not be accurate.⁶ In a previous cross-sectional study, we assessed prehospital time in patients with stroke in Berlin based on in-depth interviews.¹² In this study, patients arrived within a median time interval of 151 minutes. However, only patients with less severe states of stroke who could be interviewed were included, whereas in the present study we included all patients with stroke. With regard to sample size, the initial assumption of a 40% reduction in prehospital time intervals was later considered to be too optimistic by the study team. We therefore had to overrecruit patients. Also, we have no information about the use of the EMS because this is not routinely recorded in the medical records. Finally, we do not have information about any “adverse events” in our intervention group. For example, we do not know whether the percentage of patients admitted with suspected but unconfirmed stroke increased in the intervention areas compared with the control areas. A higher number of nonevents might increase the work burden for emergency department personnel.

In conclusion, a population-based intervention can effectively reduce prehospital delays in women with stroke by 27% but appears to have no significant effect in men. In the future, a number of issues need to be addressed such as sustainability, cost-effectiveness, transferability to other settings, clinical outcomes, and the development of sex-specific programs addressing male patients with stroke. As we find increasing evidence for the virtues of early stroke intervention, reducing prehospital delays will eventually lead to improvements in public health.

Correspondence: Jacqueline Müller-Nordhorn, MD, DPH, Berlin School of Public Health, Charité University Medical Center, Oudenarder Str 16, D-13347 Berlin, Germany (jacqueline.mueller-nordhorn@charite.de).

Accepted for Publication: May 17, 2009.

Author Contributions: Dr Wegscheider had full access to all study data and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Müller-Nordhorn, Nolte, Jungehülsing, Rossnagel, Villringer, and Willich. Acquisition of data: Nolte, Jungehülsing, Rossnagel, and Reich. Analysis and interpretation of data: Müller-Nordhorn, Wegscheider, Reich, Roll, and Willich. Drafting of the manuscript: Müller-Nordhorn. Critical revision of the manuscript for important intellectual content: Wegscheider, Nolte, Jungehülsing, Rossnagel, Reich, Roll, Villringer, and Willich. Statistical analysis: Wegscheider and Roll. Obtained funding: Nolte, Rossnagel, Villringer, and Willich. Administrative, technical, and material support: Nolte, Jungehülsing, Rossnagel, and Reich. Study supervision: Müller-Nordhorn, Nolte, Jungehülsing, Villringer, and Willich.

Financial Disclosure: None reported.

Funding/Support: This study was supported by grant 01GI9902/4 from the German Federal Ministry of Education and Research under the title “Competence Net Stroke.”

Previous Presentations: This study was presented orally at the European Society of Cardiology Congress 2008; January 9, 2008; Munich, Germany; the International Epidemiological Association (IEA) XVIII World Congress of Epidemiology; September 22, 2008; Porto Alegre, Brazil; and the Jahrestagung der Deutschen Gesellschaft für Epidemiologie; September 26, 2008; Bielefeld, Germany.

Additional Contributions: Doreen McBride, Dr.rer.medic, and Thomas Keil, MD, edited the manuscript and provided valuable advice. Annette Wagner performed an excellent job with regard to study coordination and data management.