Impact on Prehospital Delay of a Stroke Preparedness Campaign
A SW-RCT (Stepped-Wedge Cluster Randomized Controlled Trial)
Background and Purpose—Public campaigns to increase stroke preparedness have been tested in different contexts, showing contradictory results. We evaluated the effectiveness of a stroke campaign, designed specifically for the Italian population in reducing prehospital delay.
Methods—According to an SW-RCT (Stepped-Wedge Cluster Randomized Controlled Trial) design, the campaign was launched in 4 provinces in the northern part of the region Emilia Romagna at 3-month intervals in randomized sequence. The units of analysis were the patients admitted to hospital, with stroke and transient ischemic attack, over a time period of 15 months, beginning 3 months before the intervention was launched in the first province to allow for baseline data collection. The proportion of early arrivals (within 2 hours of symptom onset) was the primary outcome. Thrombolysis rate and some behavioral end points were the secondary outcomes. Data were analyzed using a fixed-effect model, adjusting for cluster and time trends.
Results—We enrolled 1622 patients, 912 exposed and 710 nonexposed to the campaign. The proportion of early access was nonsignificantly lower in exposed patients (354 [38.8%] versus 315 [44.4%]; adjusted odds ratio, 0.81; 95% confidence interval, 0.60–1.08; P=0.15). As for secondary end points, an increase was found for stroke recognition, which approximated but did not reach statistical significance (P=0.07).
Conclusions—Our campaign was not effective in reducing prehospital delay. Even if some limitations of the intervention, mainly in terms of duration, are taken into account, our study demonstrates that new communication strategies should be tested before large-scale implementation.
Stroke preparedness, meaning the ability of patients and bystanders to recognize stroke symptoms and take immediate action to seek emergency treatment,1 is among the predictors of prehospital delay,2–4 which plays a critical role in acute stroke management.5
Public education campaigns to increase stroke preparedness have been evaluated in several studies, with inconsistent and inconclusive results.6,7 Overall, stroke warning campaigns can increase the recognition of stroke symptoms, but their efficacy on patient behavior remains unproven. Although positive intervention effects are reported in the majority of studies, some methodological weaknesses, mainly in terms of design, limit the validity of the observed effects. Besides, the theoretical basis of the intervention and some exploratory work for the campaign development are not always reported, although that is recommended for the design of complex interventions.8
Two studies9,10 incorporating a sufficiently rigorous design (cluster randomized in one case and quasiexperimental in the other) demonstrated some benefit of different strategies, such as an educational letter mailed to the households9 and a multilevel campaign, developed according to a strict methodology and largely using mass media.10
The Italian Educazione e Ritardo di Ospedalizzazione per Ictus project had the primary objective of developing an educational campaign focused on stroke preparedness and aimed to reduce prehospital delay. In this article, we present the results of the evaluation trial performed to assess the intervention effectiveness. At variance with previous reports, the campaign was specifically designed for the local context according to an exploratory analysis and tested according to a rigorous design.
We used an SW-RCT (Stepped-Wedge Cluster Randomized Controlled Trial) design with cross-sectional data.11–13 The choice of the stepped-wedge design mainly relied on the available evidence of some effectiveness of previous campaigns in increasing symptom recognition, which made it unethical to withhold the intervention from a proportion of the participants. Clusters were the communities of each of the 4 provinces of Northern Emilia Romagna (Parma, Piacenza, Modena, and Reggio Emilia), ranging between 288 000 and 702 000 inhabitants (www.istat.it/en/emilia-romagna/data). According to the cross-sectional setting, the units of analysis were patients aged ≥18 years consecutively admitted to the 6 stroke receiving centers of the participating provinces for suspected stroke or transient ischemic attack. Informed consent to participate in the study was obtained from the patient or one of his/her relatives in case of severe impairment as a consequence of stroke.
Exclusion criteria were no information available about the time of stroke onset and no informed consent from patient/proxy.
The intervention was targeted at cluster level and consisted of a community campaign designed according to the Intervention Mapping framework.14
An extensive description of the campaign development is shown in the online-only Data Supplement.
In brief, as theoretical foundation, we used the General Model of Total Patient Delay15,16 and the common sense model of self-regulation.17 The message content described the most frequent symptoms, emphasized the need for calling the emergency telephone number immediately, and the availability of therapies that can lead to a complete recovery, provided that they are administered early enough. The message was related in the narrative mode, in cartoon form. Many educational products were produced: a brochure that was mailed to the households and distributed in public places, a poster (Figures I and II in the online-only Data Supplement), an animation video for closed circuit, and an animation video clip for television broadcasting.
According to the stepped-wedge design, the campaign was launched sequentially in the 4 provinces with 3-month intervals, so that its duration and intensity were not the same in the 4 clusters, lasting for a maximum of 12 months in the first province and a minimum of 3 months in the last province exposed to the intervention (Figure III in the online-only Data Supplement).
According to the campaign planning, ≈751 000 brochures depicting the comic strip were mailed to the households; additional 312 000 brochures and 400 posters depicting the Super-Hero were displayed in public places. Educational products were also distributed during 26 public meetings on a monthly basis. The closed-circuit animation video was shown during the campaign in the ED waiting rooms of the 4 major participating hospitals, starting 1 month after the campaign launch, for at least 30 days. The animation video clip on the 4 local TV stations was broadcast during the final month of the campaign a total of 385× (a detailed description of the campaign implementation across the 4 clusters is given in Table VI in the online-only Data Supplement).
The order in which the participating communities received the intervention was determined as a randomized sequence generated electronically by the Study Coordinating Center in Parma.
The intervention was compared with the usual care, meaning the spontaneous initiatives to increase stroke awareness that are usually launched by the National Health Service and Patients’ Organizations (online-only Data Supplement).
The cases were identified prospectively and on a daily basis by trained assessors who had access to the administrative data of the emergency department and to all patient medical records during in-hospital stay. Within 72 hours from hospital admission, a semistructured interview was administered to the patient or his/her caregiver, including questions about their behavior at symptom onset. The interview format is a modified version of a published instrument18 (online-only Data Supplement). If patients were unable to answer because of aphasia, motor impairment, or other stroke manifestations, any available relatives and any witnesses of stroke onset were interviewed.
The following data were recorded into an electronic Case Report Form: demographics, symptoms at onset, time of symptom onset defined as the time a neurological deficit was first noticed by the patient or an observer; time of patient presentation to the hospital ED, as recorded in the medical chart; and clinical characteristics, including scores at the National Institutes of Health Stroke Scale at hospital admission (the complete list of clinical characteristics is reported in the online-only Data Supplement).
When symptom onset was reported 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.
The primary end point was the percentage of patients who arrived at the emergency department within 2 hours from symptom onset. The 2-hour cutoff was chosen because it is the maximum prehospital delay that allows for eligible patients to receive thrombolysis within 3 hours after stroke onset, and, at the time this study was designed, the extension of thrombolysis time windows to 4.5 hours was not recommended for patients over 80 years old, who usually represent a substantial proportion of the stroke population.
Secondary end points were the proportion of all cerebrovascular patients together treated with intravenous r-tPA (recombinant tissue-type plasminogen activator) and the proportion only of patients with ischemic stroke treated with intravenous r-tPA. Furthermore, 4 behavioral end points were analyzed: the proportion of patients/proxies who attributed the symptoms to stroke; the proportion of patients who called the Emergency Services as first reaction; the proportion of persons to whom patients first referred who suggested calling the Emergency Service; and the proportion of patients who arrived with the ambulance.
Sample size was estimated taking into account that in our context, 30% of stroke patients arrive at the hospital within 2 hours from symptom onset. According to data available from literature, the proportion of patients arriving in time who do not show contraindications to treatment and can eventually be selected for thrombolysis ranges from 24% to 57%19,20; the actual and definitive treatment rate, however, depends on the efficiency of in-hospital Stroke Care organizations. So, we assumed that 40% of patients arriving in time should be eligible for treatment. On the basis of such assumption, a 15% increase of early presentation rate would translate into a 6% increase in thrombolysis rate, which in our context might reach the 24% of all ischemic strokes, starting from a baseline rate of 18%, as assessed during the 3-month baseline data collection in the 4 clusters. On these premises, the total estimated sample size was 326 U of analysis (α=0.05; power, 0.80).
Adjustment for the design was made according to Hemming and Taljaard,12 setting as design constraints a fixed number of clusters4 and of steps.5 Assuming an intracluster correlation coefficient of 0.02, the sample size increased to 960 cases (240 per cluster). The assumptions of the estimate, including the intracluster correlation coefficient value, were verified by a preliminary analysis of prehospital delay within the 4 participating hospitals for a period of 3 months. According to administrative data about the activity volume of the participating hospitals, 5 treatment periods of 3 months were deemed sufficient to achieve the sample size in all clusters.
The data were stored and analyzed in the Parma coordinating center using the Statistical Package for the Social Sciences (SPSS version 18.0; SPSS Inc, Chicago, IL) and Stata 10.0 (Stata Corp, College Station, TX).
Data cleaning was performed via SPSS syntax operations. All statistical tests were done 2-tailed with 95% confidence intervals (CIs).
The primary analysis aimed to compare stroke patients before and after the campaign implementation, according to the stepped-wedge schedule, and adjusting for clustering within communities and temporal trends.
At the patient level, the primary outcome was binary (yes, no), and so logistic regression models with binary outcomes were used. To adjust for differences in the average level of the outcome across cluster and secular trend, we applied a fixed-effect model, which has been proven as more powerful than generalized estimation equations when the number of clusters is small.21
The model incorporated intervention status as the main effect, calendar time as a continuous measure and the clustering effect, that is, effect of communities. Where appropriate, individual-level covariates and any cluster-level covariates strongly correlated with the outcome were also included in the model, to adjust for any potential confounding. Covariates were selected on the basis of their clinical relevance to the outcome of interest as reported from other studies and their significance in univariate regression analysis (P<0.15).
The estimated intervention effect was reported as odds ratio (OR) and was considered significant at the 5% level.
Analysis of the secondary outcomes took a similar form to that described for the primary outcome. For each outcome, the intracluster correlation coefficient was estimated by 1-way ANOVAs.
Finally, a sensitivity analysis was performed, to assess the robustness of the missing data assumption made in the primary analysis and to test whether the study results were influenced by factors such as the inclusion of cases with nonexact time of onset and the assumption that the 2-hour threshold was the most appropriate to categorize the delay. A complete description of sensitivity analysis is given in the online-only Data Supplement.
This study has received individual Research Ethics Board approval from all 4 of the participating provinces in 2012.
The trial start and finish dates were prespecified as August 1, 2013 to November 30, 2014. Figure 1 shows a diagram depicting the rollout of the campaign within the 4 clusters in the 5 periods and the number of analysis units enrolled within each cluster in each study period. During the 15-month study period, 1714 patients were enrolled as analysis units. In 27 cases, the onset time as recorded in the case report form was not congruous with the time of hospital arrival, and in 65 National Institutes of Health Stroke Scale scores were missing. Thus, a complete case analysis was performed on a data set of 1622 patients, 912 exposed and 710 nonexposed (Figure 2).
Demographic and clinical characteristics of all participants and according to trial mode are represented in Table 1. A higher proportion of patients with more severe stroke and with a different distribution of ischemic stroke causes were enrolled during the intervention period.
The comparison between clusters, which showed some differences, especially in stroke severity at admission, is shown in Table III in the online-only Data Supplement.
The median (interquartile range) time from symptom onset/awareness to presentation at the hospital in the whole sample was 2 hours and 40 minutes (1 hour and 22 minutes to 11 hours and 16 minutes). Six hundred sixty-nine patients (41.2%) presented <2 hours after stroke onset/symptom awareness.
According to univariate regression analysis (Table IV in the online-only Data Supplement), older age, living in urban areas, previous stroke or transient ischemic attack, atrial fibrillation, diagnosis of transient ischemic attack, dyslipidemia, coronary heart disease, higher severity of neurological impairment at onset, and cardioembolic cause were predictive of early hospital admission, whereas male sex, living alone, diabetes mellitus, smoking habits, and symptom onset during night hours and at awakening were associated with later arrival. A significant association with prehospital delay was also found for cluster, with cluster 4 patients arriving earlier.
The median (interquartile range) time interval was 2 hours and 59 minutes (1 hour and 26 minutes to 13 hours and 8 minutes) during the campaign exposure, and 2 hours and 26 minutes (1 hour and 59 minutes to 8 hours and 59 minutes) during nonexposure period.
Table 2 shows the results of the analysis of the campaign effect on early arrival and thrombolysis rate. As for early arrival, 3 different models were developed: the first, adjusting for cluster and time; the second, adjusting also for National Institutes of Health Stroke Scale score and age (the potential confounders); and the third, adjusting also for the covariates variables previously identified by univariate analysis as determinants of the delay. As for thrombolysis, adjustment was made for National Institutes of Health Stroke Scale and age, which are 2 main criteria that are taken into account for patient selection.
The proportion of patients who arrived within 2 hours of stroke onset was lower during the campaign (354 [38.8%] versus 315 [44.4%]), but the effect estimates according to the ORs were not significant both before (OR, 0.86; 95% confidence interval [CI], 0.66–1.14; P=0.29) and after adjustment for potential confounders (OR, 0.84; 95% CI, 0.63–1.11; P=0.23) and other determinants of delay (OR, 0.81; 95% CI, 0.60–1.08; P=0.15).
Thrombolysis rate in the whole sample was 19.1%, with a proportion of 24.5% of ischemic strokes. The rate was lower during the campaign, but the difference in patients with ischemic stroke was not significant both in unadjusted and adjusted analysis.
As for behavioral end points (Table 3), exposure to the campaign was associated with an increase of the proportion of stroke recognition (from 26.3% to 34.6%) with an OR in the logistic regression of 1.48 (95% CI, 1.19–1.84; P<0.001). According to the fixed-effect model, adjusting for cluster and time, the positive effect estimate was confirmed, but with a loss of significance (OR, 1.40; 95% CI, 0.96–2.03; P=0.07). No significant change was found for the other behavioral end points, including calling 118 and hospital arrival with an ambulance.
Sensitivity analyses (Table V in the online-only Data Supplement) performed on multiple imputed data sets, after exclusion of cases with nonexact onset time, using different cutoffs for prehospital delay definition and according to a shared frailty model for time to event data, showed similar results.
As for process analysis, the intervention was delivered, according to the protocol, without any significant variations, across the 4 clusters (online-only Data Supplement).
In this study, a public educational campaign aimed at increasing stroke awareness and preparedness was not effective in reducing prehospital time intervals, neither was it effective in increasing the rate of thrombolysis for ischemic stroke. On the contrary, it was associated with a nonsignificant decline in early arrival. The lack of effectiveness was accounted for by the lack of beneficial effects on behavioral end points, except for the slight and nonsignificant increase in stroke recognition.
Several factors might explain the campaign failure.23 An inadequate implementation of the campaign is unlikely because the process evaluation ensured the homogeneous delivery of the intervention across clusters, except for the differences in campaign duration according to the SW design.
The possibility of some kind of interference by other contemporary campaigns on stroke preparedness, conducted by other organizations, which might have contributed to our intervention failure, can be excluded as well. As stated in the online-only Data Supplement, the usual care that served as comparison in our study did not include any specific and organized intervention to increase stroke preparedness throughout the study period. There were only occasional meetings with community groups, with distribution of educational brochures on stroke and screening interventions to identify high-risk individuals, which are usually organized once a year by the Italian Association for Fighting Stroke (Associazione per la Lotta all’Ictus Cerebrale).
Thus, other reasons of poor performance should be considered, such as limitations of the theoretical foundation of the intervention and of its components, notably in terms of mode, appeal, and communication channels.23
As for the appeal, which is the way of organizing the content of the message to make it more likely to persuade or convince people,24 it cannot be excluded that emotional or fear-arousal appeals might be more effective than the more rational and positive appeal we used. However, the UK Act FAST campaign, which used a fear-arousing appeal, depicting stroke onset as a fire spreading in a TV advertisement,25–27 produced contradictory results.
Nevertheless, our study is consistent with the evidence from previous reports, which did not find a significant reduction of prehospital delay after exposure to public education campaigns on stroke,25,26,28–31 even when multilevel interventions, using mass media, were evaluated.10 One exception is the reduction of prehospital delay reported for women by Müller-Nordhorn et al.9
Many study limitations should be taken into account. The small number of clusters (4) might represent a critical issue because a minimum of 10 clusters is recommended to be used in each arm of a cluster randomized trial.33 Unfortunately, economic constraints prevented the involvement of more Italian provinces into the project. However, the fixed-effect regression model performs quite well in modeling clustered data with very few clusters.21,34 Besides, the stepped-wedge design may mitigate some of the potential issues surrounding the small number of clusters, making the magnitudes of intracluster correlations less important.35 Furthermore, methodological constraints affected the selection of channels for message delivery. The risk of contamination between clusters (the 4 provinces) because of the potential overlapping of local media orbits limited the use of internet and television public service announcements. Most of all, the length of the intervention might be a critical aspect.23 In our study, the exposure periods ranged between 3 months in the fourth and 12 months in the first cluster. In this view, the improvement of stroke recognition, although not significant, might be considered a promising finding which should be viewed as preliminary to further interventions, although the economic sustainability of campaigns of longest duration may represent a major issue.
Finally, the lack of improvement in thrombolysis rate deserves some considerations. In our study, the rate of thrombolysis was rather high, reaching the 25% of ischemic stroke patients (18% in the 3-month baseline period). This finding is not unexpected, since the last few years have witnessed a rise in both the diffusion of Stroke Units and access to thrombolytic therapy in Italy, in general and in our region in particular, as the consequence of the beneficial policy of the Emilia Romagna Health System of supporting the implementation of Stroke Units in its major hospitals. Nevertheless, there is place for improvement because, according to a 2013 report of our regional Health System, in Emilia Romagna, still 50.3% of patients eligible for thrombolysis remained untreated.
Anyway, because of the high rate of thrombolysis at the baseline already, only little change was likely to be found, which introduces a type of ceiling effect, allowing only minor changes as a result of the intervention. In this context, a larger sample size is needed to achieve statistically significant results.
In conclusion, our campaign was not effective in reducing prehospital delay of stroke patients, although some limitations of the intervention, mainly in terms of duration, could explain its failure. Overall, the study demonstrates that any new communication strategies, even if rigorously designed, should be properly tested before large-scale implementation.
We wish to thank the organization Associazione per la Lotta all’Ictus Cerebrale for contributing to the campaign development.
Sources of Funding
This work received grants from the Programma di Ricerca Regione Università Emilia Romagna 2011 to 2012 AREA 2 Ricerca per il governo clinico.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.018135/-/DC1.
- Received June 13, 2017.
- Revision received August 20, 2017.
- Accepted September 18, 2017.
- © 2017 American Heart Association, Inc.
- Boden-Albala B,
- Quarles LW
- Teuschl Y,
- Brainin M
- Emberson J,
- Lees KR,
- Lyden P,
- Blackwell L,
- Albers G,
- Bluhmki E,
- et al
- Morgenstern LB,
- Staub L,
- Chan W,
- Wein TH,
- Bartholomew LK,
- King M,
- et al
- Hemming K,
- Haines TP,
- Chilton PJ,
- Girling AJ,
- Lilford RJ
- Campbell MK,
- Piaggio G,
- Elbourne DR,
- Altman DG
- Bartholomew LK,
- Markham CM,
- Ruiter RAC,
- Fernandez ME,
- Kok G,
- Parcel GS
- Cameron LD,
- Leventhal HA
- Leventhal HA,
- Brissette I,
- Leventhal EA
- Hsia AW,
- Castle A,
- Wing JJ,
- Edwards DF,
- Brown NC,
- Higgins TM,
- et al
- McNeish D,
- Stapetlon LM
- Addo J,
- Ayis S,
- Leon J,
- Rudd AG,
- McKevitt C,
- Wolfe CD
- Dombrowski SU,
- Mackintosh JE,
- Sniehotta FF,
- Araujo-Soares V,
- Rodgers H,
- Thomson RG,
- et al
- 28.↵NINDS. Know Stroke Campaign. https://stroke.nih.gov/. Accessed June 28, 2016.
- 29.↵Power to End Stroke. http://www.empoweredtoserve.org/index.php/power-to-end-stroke/. Accessed July 7, 2016.
- Boden-Albala B,
- Edwards DF,
- St Clair S,
- Wing JJ,
- Fernandez S,
- Gibbons MC,
- et al
- 33.↵Medical Research Council Clinical Trials Series. Cluster Randomised Trials: Methodological and Ethical Considerations. London: MRC;2002. https://www.cebma.org/wp-content/uploads/Cluster-randomised-trials-Methodological-and-ethical-considerations.pdf. Accessed October 26, 2017.