Article Text
Abstract
Background Inhospital stroke (IHS) is associated with high morbidity and mortality, likely related to multiple factors, including delayed time to recognition, associated comorbidities, and initial care from non-stroke trained providers. We hypothesized that guided revision of a formalized ‘stroke code’ system can improve diagnosis and time to thrombolysis and thrombectomy.
Methods IHS activations occurring at a comprehensive stroke center between 2013 and 2016 were retrospectively analyzed to guide revisions of an established stroke code protocol to improve provider communication and time to imaging, reduce stroke mimic rate, and improve the use of parallel processing. After protocol implementation, we prospectively collected data between 2016 and 2017 for comparison with the pre-implementation group, including diagnostic accuracy and relevant time points (code call to examination, examination to imaging, and imaging to intervention). We report descriptive statistics for comparison of patient characteristics and time metrics (time to imaging and reperfusion after IHS activation). Multivariable regression analysis was performed to identify independent predictors of stroke mimics and time metrics.
Results There were 136 cases in the pre-implementation group and 69 in the post-implementation group. A reduction in stroke mimics (52% vs 33%, P=0.01) occurred after protocol initiation. Mean time to imaging after stroke code call was 7.6 min shorter (P=0.026) and mean time from imaging to acute reperfusion therapy was 45.7 vs 19.8 min (P=0.05) in the pre- versus the post-implementation group.
Conclusion Revision of an existing IHS protocol was associated with a lower rate of stroke mimics, and a shorter time to intravenous and intra-arterial intervention.
- stroke
- thrombectomy
- thrombolysis
Statistics from Altmetric.com
Introduction
Patients developing inhospital strokes (IHS) have higher morbidity and mortality compared with patients with community onset stroke presenting to the emergency department,1–3 despite ready access to providers, diagnostic testing, and treatments. Worse outcomes after IHS are multifactorial, including risk factors inherent to inpatient status and delays in evaluation and treatment.4–7 Most inpatients are admitted to non-neurological services when they develop a stroke, and clinical staff may not be accustomed to rapid evaluation of stroke symptoms.8 9 Management of IHS is further complicated by high rates of mimics, estimated to be almost half of IHS activations in one multicenter evaluation.10–12 Use of thrombolysis is limited by delayed recognition and other contraindications (eg, coagulopathy, recent surgery, etc), which may also increase morbidity compared with out of hospital stroke cases.13
Overcoming unique obstacles to diagnosis and treatment of IHS is an area of ongoing research. Prior work demonstrated the benefits of an established inpatient stroke team, protocols for stroke code response, and bundled evaluations to reduce inhospital delays.14–18 Sequential time metrics improved with use of computer based physician order entry, flagging potential reperfusion candidates, and improving communication between departments, although this was limited to patients admitted to a cardiology ward and analysis did not include patients who did not undergo reperfusion.16Little is known regarding specific sources of delay and how to target workflow to deliver faster treatment and reduce morbidity. One potential strategy to improve efficiency is implementation of parallel workflow, which has been shown to reduce door to groin times in the emergency room setting.19 In this study, we sought to understand sources of delay in IHS within an established inpatient stroke code protocol at a large academic campus. We then implemented guided protocol modifications to include parallel processing to improve IHS recognition, assessment, and treatment, and compared data points of interest with the pre-implementation cohort.
Methods
Study setting
Evaluation of IHS cases occurred at a comprehensive stroke center, with adjoining inpatient ward and outpatient clinic settings within a single campus. At our institution, a ‘stroke code’ activation occurs via overhead hospital wide announcement and a direct page to the neurology resident led stroke team; both notifications include the reported patient location. Per prior protocol structure, most cases of a stroke code are preceded by involvement of a rapid response team led by critical care trained providers, through a ‘condition C,’ which is communicated in a similar fashion as stroke codes. Both types of activation involve the relevant response team, including physicians, nurses, and administrators to facilitate patient transfers.
A retrospective analysis of stroke code activations for patients on non-neurology services (unknown to the stroke team) between 2013 and 2016 was completed to inform guided protocol modifications to an established stroke code protocol. Review of cases in the pre-implementation cohort showed a high rate of stroke code activation for stroke mimics. Workflow of IHS codes during this period was also notable for inefficient communication between the rapid response and neurology teams without a means of direct contact, and without feedback or review of individual patient outcomes by those involved. Analysis of relevant time points identified delays in time to imaging and treatment.
Protocol implementation
Key stakeholders involved in stroke code activations (nursing, critical care medicine, neurology, interventional neurology, and radiology) participated in an interdisciplinary effort to develop a revised IHS protocol (figure 1). The protocol incorporated multiple changes to improve workflow, as described below, and was combined with stroke education for rapid response providers. We specifically intended to replicate the process used for community onset strokes presenting to an emergency department, with rapid imaging and simultaneous involvement of providers (eg, simultaneous data gathering, IV placement, examination, etc). Institutional review board approval was obtained prior to data collection, and all measures were supported by stakeholders and approved through a local quality improvement committee prior to initiation. Implementation occurred after a 3 month lead-in period with prospective data collection between October 2016 and December 2017.
Steps to improve stroke recognition
Given most initial assessments were completed by non-neurology trained providers as part of the rapid response team, dedicated stroke education to critical care fellows was completed prior to the initiation of the revised protocol. Formal training included stroke symptoms and treatment algorithms of ischemic and hemorrhagic stroke through a dedicated lecture series (Emergency Neurologic Life Support course developed by the Neuro-critical Care Society). Review and demonstration of a rapid stroke symptom severity scale (RACE, rapid arterial oCclusion Evaluation) was also completed to aid in identification of focal neurologic deficits. RACE was selected to heighten sensitivity for large vessel occlusion and given its ease of use for non-neurology trained providers.
Steps to improve communication
Direct communication between the rapid response team and neurology was mediated through a dedicated telephone carried by the stroke team. A detailed graphical version of the revised protocol was created and provided to all stakeholders with additional availability of an electronic version for real-time review.
Steps to improve time to imaging and therapy
The protocol highlighted early notification of the neurology service for suspected strokes and parallel processing (eg, concurrent hemodynamic assessment and glucose check, patient transport to imaging (either CT or MRI) and communication with radiology similar to evaluations of community onset stroke equivalents (see figure 1)). An emphasis on rapid imaging included use of a ‘CT rendezvous’ paradigm in which transport of patients to the scanner occurred at the direction of the rapid response team, allowing the neurology team to meet the patient directly in the radiology department for completion of the examination and review of the clinical information.13 Further steps included the acceptance of verbal orders for imaging by the radiology department, and an expectation for therapeutic decision making in all cases to occur prior to patient transport out of the radiology department (to either return to the original patient unit or transfer to a stroke unit). Of note, beginning in the pre-implementation cohort in December 2014, intravenous tissue plasminogen activator (IV tPA) was brought to the bedside for all IHS codes as part of a protocol amendment. The practice of bringing IV tPA to stroke code activations continued in the post-implementation cohort with a provision for direct dispatch to the radiology department if needed.
Feedback
Rolling feedback was provided for all IHS cases in the post-implementation period (2016–2017) to individuals of the rapid response and neurology teams. Feedback via electronic mail notification included a brief description of the stroke code evaluation, measured time points and outcomes, with additional directed questions regarding workflow, when appropriate, and reinforcement of elements of the protocol.
Data collection
Analysis for all IHS code activations was completed for two patient cohorts: retrospectively for patients between March 2013 and August 2016, and prospectively for patients between October 2016 and December 2017 after protocol revision. We identified patients by querying a prospectively created local ’Get with the Guidelines' database, which included any instance of stroke code activation (including strokes and stroke mimics). We excluded patients already admitted to the inpatient stroke service.
We collected additional data through electronic medical record review, including: patient demographics; baseline medical comorbidities; clinical variables at the time of stroke code, including documented National Institutes of Health Stroke Scale (NIHSS), blood pressure, international normalized ratio (INR), creatinine, and glucose; and discharge disposition. We compared the following time points between both groups: last known well to symptom recognition, stroke code call to neurology bedside assessment, stroke team assessment to imaging, imaging to IV tPA, and imaging to groin access for thrombectomy. We also recorded the patient location and identified the primary service to whom the patient was admitted.
Statistical analysis
Continuous variables are reported as mean (±SD) or median (IQR), as appropriate. Categorical variables are reported as proportions. Study patients were divided into pre- and post-implementation groups. The primary objectives included incidence of stroke mimics, time to imaging, and time to acute reperfusion therapy (time to first treatment for each patient, either thrombolysis or thrombectomy). Between group comparisons for continuous variables were performed using the Student’s t test and for categorical variables using the χ2 or Fisher exact test, as appropriate. Univariable analysis and multivariable logistic regression analysis was performed to identify associations after adjusting for known confounders, including age, sex, NIHSS score, surgical history, anesthesia characteristics, vascular risk factors, and history of migraine. Associations are presented as OR (95% CIs). Significance was defined as P≤0.05. Statistical analysis was performed using IBM SPSS Statistics 23 (IBM-Armonk, New York, USA).
Results
Patient demographics
There were 205 patients admitted to non-neurology services included for analysis in whom a stroke code was activated: 136 in the pre-implementation phase (2013–2016) and 69 in the post-implementation phase (2016–2017). Patient characteristics were similar between the pre- and post-implementation groups (table 1), including age (mean 67±14.1 vs 66±14.3 years), gender (47% vs 46% women), and stroke severity (mean NIHSS 9 vs 10). Measures of systolic and diastolic blood pressure at the time of the stroke code call, mean INR, antiplatelet and anticoagulant use, periprocedural stroke codes, and baseline medication conditions were also similar between the groups (table 1).
Location of stroke code and diagnoses of stroke mimics
IHS cases occurred in both the inpatient and outpatient (clinic) settings within the hospital campus. Most cases occurred on inpatient wards, with only a fraction occurring in an intensive care unit: 12.5% and 27.5% occurring in an intensive care unit in the pre- and post-implementation cohorts, respectively. In the post-implementation cohort, more cases occurred in patients admitted to a cardiology service than prior to initiation of the revised protocol (42.1% vs 23.5%).
The overall rate of stroke mimics (table 2) significantly decreased after revisions to the stroke code protocol (52.9% vs 33.3%, P=0.01). The most common etiologies of stroke mimics in both cohorts were encephalopathy and seizure. The rate of stroke mimics also varied based on admission location and group. Rates of stroke mimics significantly decreased in patients admitted to a primary inpatient medicine service after protocol implementation (71.4% (25/35) vs 33.3% (4/12), P=0.001). Patients admitted to a cardiology unit trended toward reduced mimic rates in the post-implementation phase (59.3% (19/32) vs 37.9% (11/29), P=0.09). Multivariable logistic regression using pre- and post-implementation groups as a binary variable and including variables with a P value <0.2 in univariable analysis, did not identify any independent predictors of stroke mimics (table 3).
Time metrics
Chart review of both groups found that time of last seen well was limited in perioperative cases, as well as in those involving sedating medications. Mean assessment and treatment times in the pre- and post-implementation cohorts were: last known well to symptom recognition (177.4 min vs 180.6 min, P=0.938), stroke code call to stroke team assessment (9.7 min vs 5.1 min, P=0.011), stroke code call to imaging (40 min vs 32.4 min, P=0.026), and imaging to acute reperfusion therapy (either thrombolysis or thrombectomy) (45.7 min vs 19.8 min, P=0.05) (see figure 2). Regression analysis for time to imaging did not identify any independent predictors.
Treatment
In the pre-implementation cohort, 48 patients were identified clinically as stroke out of 136 IHS codes. Of these, IV tPA was administered in 12 patients and intra-arterial thrombectomy was performed in 14 patients (3 patients received both). Despite improvement in nearly all time points, rates of reperfusion therapies did not increase in the post-implementation cohort, and in fact decreased overall. IV tPA was given in 3 patients, and intra-arterial thrombectomy was performed in 6 patients, of a total of 36 patients identified with ischemic stroke/transient ischemic attack (3 patients received both) in the post-implementation cohort.
Contraindications to intravenous thrombolysis in the pre- and post-implementation cohorts were: rapidly improving or minimal deficits (28.6% vs 21.2%); recent surgery (23.8% vs 21.2%); outside the tPA time window (21.4% vs 33.3%); coagulopathy (19.0% vs 15.2%); and recent stroke (7.1% vs 9.1%). Ineligibility for tPA occurred in 75% of patients ultimately diagnosed with stroke or transient ischemic attack in the pre-implementation cohort, and in 92% of patients in the post-implementation cohort.
Thrombectomy was performed in 87% of large vessel occlusions in the pre-implementation cohort, and in 66% of large vessel occlusions in the post-implementation. Thrombectomy was performed in all patients with large vessel occlusion with the following exceptions: in the pre-implementation group, two patients recanalized after IV tPA, and in one patient without a clinical/imaging mismatch and poor Alberta Stroke Program Early CT Score; in the post-implementation group, two patients had low symptom severity (NIHSS <4) and two patients lacked a clinical/imaging mismatch (poor Alberta Stroke Program Early CT Score) to warrant intervention.
Outcome
Mortality and discharge disposition were analyzed in both groups, and were as follows in patients with stroke/transient ischemic attack in the pre-implementation cohort versus the post-implementation cohort: death occurred in 8.3% vs 16.6% (P=0.24), discharge to home in 37.5% vs 22.2% (P=0.13).
Discussion
By revising a pre-existing IHS code protocol, we were able to reduce the frequency of code calls for stroke mimics and substantially reduce recognition to imaging and treatment times. We first completed a retrospective review of the established stroke code process in our institution which was notable for poor inter-provider communication, sources of delay in evaluation (time to imaging) and treatment (time to thrombolysis, endovascular therapy), as well as high rates of activations for stroke mimics. The information gained from the analysis was used to guide a quality improvement project, including multidisciplinary protocol revision of stroke code activations, ability for early provider communication during activations, and dedicated education for non-neurology trained providers. Protocol revision included parallel processing when able to reduce time of evaluation and treatment.
We used a phased lead-in period to allow providers to become familiar with the protocol and troubleshoot any areas of limitation. This period was particularly helpful given rotating call periods for rapid response and neurology providers and variable occurrence of stroke codes on any given day. Additionally, we found provision of an electronic version of the protocol, RACE score, and radiology contact information were useful additions during the lead-in period to aid in triage and communication moving forward.
After implementation, we noted significant reductions in time to neurologic assessment, imaging, and in reperfusion therapy. It is possible that at least some improvement in time metrics occurred due to the act of measurement itself (ie, the Hawthorne effect), although rolling feedback occurred after the completion of each stroke code and the providers involved varied throughout the study period. Additionally, use of standardized protocols were shown to reduce time to thrombectomy, which may also play a role in our results.20 21 Regardless, the greatest reduction occurred in time from imaging to treatment, with an average reduction of 25 min. Prior estimates suggest an average time to treatment of 100 min or more, well beyond the accepted goal of 60 min, which is achieved in only one in five patients with IHS.22 23 In our post-implementation cohort, time to treatment from code call was less than the benchmark of 60 min in all three cases of tPA administration and in five of the six cases of thrombectomy (the remaining case occurred in 66 min). Further reductions in time to evaluation and treatment are possible via continued use of the established protocol, including expanded use of CT rendezvous.
We did not find that improved time to imaging and treatment translated into a greater proportion of IHS undergoing thrombolysis or thrombectomy in the post-implementation cohort. Although the stroke codes initiated during the post-implementation cohort were less likely to be for a stroke mimic, the actual proportion of reperfusion therapy decreased, which was an unexpected finding. This analysis is limited by a relatively small subset of patients treated overall in both groups and we were unable to control for potential intergroup differences in presentation. Part of this difference can be further attributed to baseline patient characteristics, with a higher rate of tPA contraindications in the post-implementation cohort.
Notably, a substantial time delay from last seen well (LSW) to the time of recognition (time of stroke code call) accounted for the greatest delay in processing. The average time from LSW to code call was approximately 3 hours in both groups, and thus contributed to reduced thrombolysis use. We did not target the time from LSW to stroke symptom recognition in this current study although this is a potentially fruitful area for future protocol modifications as widespread education has proven an effective strategy in community onset strokes.24
Similar to previous studies, we found a pattern of limited recognition of symptom onset in many cases due to confounding polypharmacy and postoperative status.8 Expansion of stroke education to additional providers could improve recognition and reduce the time between LSW and activation of the stroke code. Nursing staff were responsible for almost two thirds of symptom recognition in a prior evaluation of IHS, highlighting the need to expand education beyond those involved in the rapid response teams to include all levels of care providers.9 If effective, reductions in LSW to code call may translate to a greater and earlier use of therapeutic thrombolysis and thrombectomy.
The most common causes of stroke mimics in our cohorts were similar to those reported in earlier studies, which found highest rates for seizure, hypoperfusion, delirium, metabolic disturbance, or medication effect.11 Importantly, the post-implementation phase was associated with significantly reduced rates of stroke mimics from over half to around one-third of stroke code cases. We suspect the additional stroke symptom teaching (ENLS) and completion of training with a stroke severity scale by non-neurologists was primarily responsible for the associated reduction in stroke mimics. Additionally, the use of rolling feedback and general knowledge of our protocol likely may also have contributed to this decline. Recent published work suggests weighting patient characteristics in combination with a simplified stroke severity scale to identify cases of IHS could further reduce mimic rates.12 In general, any one provider outside of the neurological inpatient service will infrequently encounter an acute stroke, thus education of stroke signs plays an important role in identification. In this effort, we found that our targeted stroke education and rolling feedback was associated with fewer stroke mimics and a greater proportion of true stroke cases identified. We recognize that our single center protocol implementation is a potential limitation in feasibility in other institutions but the concepts within our model (dedicated training of responding providers, parallel processing, and improved communication) could be applied for future encounters across institutions of practice.
The data reviewed in this study included patients admitted to non-neurology services and not those already known to the stroke service, which is already comfortable in the evaluation and treatment of IHS. We focused on the acute management of possible stroke cases; thus the cohort also did not include IHS in which a stroke code activation did not occur. It is possible that a lower rate of stroke mimics in the post-implementation cohort represents an overall reduction in patients identified (increased false negatives) by non-neurologists and thus not included in our cohort. However, we suspect this is highly unlikely given an increase in the overall rate of stroke codes in the post-implementation cohort (3.1 vs 4.6 per month) and improved provider stroke education. Further, we suspect that persistent symptoms after the initial evaluation during a stroke code would prompt repeat neurologic evaluation. Although missed IHS are not quantifiable based on our analysis at this time, further education of non-neurology trained providers at all levels could bring more cases to recognition and, in turn, potential intervention.
Lastly, our data analysis was limited to clinical outcome measures including discharge disposition or death as a primary marker. We acknowledge that a functional status marker like the modified Rankin Scale score would be further beneficial to evaluate potential improvement in the post-implementation cohort as a result of improved treatment times. We additionally acknowledge that the effect of treatment on outcome may be difficult to ascertain given the heterogeneity of the cohort and given admission for a non-stroke diagnosis, with additional comorbidities due to the presenting illness. The presence of these confounders also limits comparisons with out of hospital strokes.
Conclusion
An IHS code protocol to include first responder stroke education, improved provider communication, and greater parallel processing can lead to significant reductions in time to evaluation and treatment, and a reduction in stroke code activations for stroke mimics. A focus of future protocol modification is to reduce time from last seen well to recognition of symptoms.
References
Footnotes
Contributors All authors listed contributed to our project as follows: JM: author design and conceptualization, acquisition of the data, analysis and interpretation of the data, and manuscript draft. NK: acquisition of the data, analysis and interpretation of the data, and revising of the article. AB: acquisition of the data, analysis and interpretation of the data, and revising of the article. YA: acquisition of the data, analysis and interpretation of the data, and revising of article. SMD: acquisition of the data, analysis and interpretation of the data, statistical analysis, and revising of the article. BM: acquisition of the data, analysis and interpretation of the data, statistical analysis, and revising of the article. WTD: acquisition of the data, analysis and interpretation of the data, and revising of the article. CM-G: acquisition of the data, analysis and interpretation of the data, and revising of article. LS: acquisition of the data, analysis and interpretation of the data, and revising of the article. TGJ: acquisition of the data, analysis and interpretation of the data, and revising of the article. APJ: design and conceptualization, acquisition of the data, analysis and interpretation of the data, and study supervision.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial, or not-for-profit sectors.
Competing interests None declared.
Ethics approval Institutional review board approval was obtained prior to data collection, and all measures were supported by stakeholders and approved through a local quality improvement committee prior to initiation of the study.
Provenance and peer review Not commissioned; externally peer reviewed.
Patient consent for publication Not required.