Accuracy of Emergency Medical Services Dispatcher and Crew Diagnosis of Stroke in Clinical Practice

Background Accurate recognition of stroke symptoms by Emergency Medical Services (EMS) is necessary for timely care of acute stroke patients. We assessed the accuracy of stroke diagnosis by EMS in clinical practice in a major US city. Methods and results Philadelphia Fire Department data were merged with data from a single comprehensive stroke center to identify patients diagnosed with stroke or TIA from 9/2009 to 10/2012. Sensitivity and positive predictive value (PPV) were calculated. Multivariable logistic regression identified variables associated with correct EMS diagnosis. There were 709 total cases, with 400 having a discharge diagnosis of stroke or TIA. EMS crew sensitivity was 57.5% and PPV was 69.1%. EMS crew identified 80.2% of strokes with National Institutes of Health Stroke Scale (NIHSS) ≥5 and symptom duration <6 h. In a multivariable model, correct EMS crew diagnosis was positively associated with NIHSS (NIHSS 5–9, OR 2.62, 95% CI 1.41–4.89; NIHSS ≥10, OR 4.56, 95% CI 2.29–9.09) and weakness (OR 2.28, 95% CI 1.35–3.85), and negatively associated with symptom duration >270 min (OR 0.41, 95% CI 0.25–0.68). EMS dispatchers identified 90 stroke cases that the EMS crew missed. EMS dispatcher or crew identified stroke with sensitivity of 80% and PPV of 50.9%, and EMS dispatcher or crew identified 90.5% of patients with NIHSS ≥5 and symptom duration <6 h. Conclusion Prehospital diagnosis of stroke has limited sensitivity, resulting in a high proportion of missed stroke cases. Dispatchers identified many strokes that EMS crews did not. Incorporating EMS dispatcher impression into regional protocols may maximize the effectiveness of hospital destination selection and pre-notification.

Background: Accurate recognition of stroke symptoms by Emergency Medical Services (EMS) is necessary for timely care of acute stroke patients. We assessed the accuracy of stroke diagnosis by EMS in clinical practice in a major US city.
Methods and results: Philadelphia Fire Department data were merged with data from a single comprehensive stroke center to identify patients diagnosed with stroke or TIA from 9/2009 to 10/2012. Sensitivity and positive predictive value (PPV) were calculated. Multivariable logistic regression identified variables associated with correct EMS diagnosis. There were 709 total cases, with 400 having a discharge diagnosis of stroke or TIA. EMS crew sensitivity was 57.5% and PPV was 69.1%. EMS crew identified 80.2% of strokes with National Institutes of Health Stroke Scale (NIHSS) ≥5 and symptom duration <6 h. In a multivariable model, correct EMS crew diagnosis was positively associated with NIHSS (NIHSS 5-9, OR 2.62, 95% CI 1.41-4.89; NIHSS ≥10, OR 4.56, 95% CI 2.29-9.09) and weakness (OR 2.28, 95% CI 1.35-3.85), and negatively associated with symptom duration >270 min (OR 0.41, 95% CI 0.25-0.68). EMS dispatchers identified 90 stroke cases that the EMS crew missed. EMS dispatcher or crew identified stroke with sensitivity of 80% and PPV of 50.9%, and EMS dispatcher or crew identified 90.5% of patients with NIHSS ≥5 and symptom duration <6 h.
conclusion: Prehospital diagnosis of stroke has limited sensitivity, resulting in a high proportion of missed stroke cases. Dispatchers identified many strokes that EMS crews did not. Incorporating EMS dispatcher impression into regional protocols may maximize the effectiveness of hospital destination selection and pre-notification.  inTrODucTiOn To be maximally effective, stroke therapies, including tissue plasminogen activator (rt-PA) and endovascular thrombectomy (ET), must be delivered as quickly as possible (1). The American Heart Association recommended development of regionalized systems of care, preferentially transporting patients to the nearest stroke center, rather than the nearest hospital (2,3). These recommendations are being adopted across the US (4). The impetus to bring patients with severe stroke directly to a Comprehensive Stroke Center is particularly pressing given randomized trials showing benefit of endovascular therapy (1). Regionalized systems of care are dependent on early and accurate identification of stroke patients by Emergency Medical Services (EMS). Although validated prehospital stroke scales exist, the diagnostic sensitivity of EMS varies from 44 to 72% in clinical practice (5)(6)(7)(8)(9). We aimed to determine prehospital diagnostic accuracy of EMS dispatchers and crews for stroke overall, for acute stroke patients with National Institutes of Health Stroke Scale (NIHSS) ≥5, and which clinical features were associated with correct prehospital identification of stroke.

MaTerials anD MeThODs
We performed a retrospective observational study using data from September 2009 to October 2012 comparing prehospital diagnosis to discharge diagnosis of patients arriving via EMS to the Hospital of the University of Pennsylvania (HUP). The study protocol was approved by the Hospital of the University of Pennsylvania Institutional Review Board.
We matched patients identified as potential stroke in the Philadelphia Fire Department (PFD) database, a repository of clinical information for every 9-1-1 EMS encounter in Philadelphia, with confirmed stroke or TIA in the HUP Get with the Guidelines (GWTG) database, using patient identifiers, location, time of call, dispatcher impression, crew impression, and patient care report narratives. The PFD is the sole 9-1-1 EMS response agency for the City of Philadelphia. Patients who did not arrive via EMS or who could not be cross-matched were excluded. We recorded admission NIHSS, symptom duration on arrival, thrombolytic treatment, and discharge disposition. Final diagnosis was classified as TIA, infarct, hemorrhage, stroke mimic, or other. Stroke mimics included: seizure, hypoglycemia, intracranial tumor, spinal cord pathology, encephalopathy, migraine, recrudescence of previous stroke symptoms, and neuropathies.
Emergency Medical Services diagnosis and discharge diagnosis were compared to determine dispatcher and crew sensitivity and positive predictive values (PPV) for all stroke and TIA patients and separately for patients with NIHSS ≥5 presenting <6 h of symptom onset, to represent the cohort most likely to be eligible for rt-PA and ET. We calculated the number of stroke or TIA patients who were correctly identified by both EMS crew and dispatcher, and by each group alone. We created a logistic regression model including age, sex, stroke type (infarct versus hemorrhagic), NIHSS (<5, 5-9, 1 ≥ 0), clinical features (weakness, speech difficulty, altered mental status, ataxia, vision loss, and neglect determined by retrospective chart review), and symptom duration (<180, 180-270, and >270 min).
In a secondary analysis, we evaluated for differences between subjects with a vascular event who were correctly identified by dispatchers, but not EMS crews, and all other subjects with a vascular event. Pearson's chi-squared test was used to examine for differences based on age, sex, stroke type (infarct versus hemorrhagic), NIHSS (<5, 5-9, 10≥), clinical features (weakness, speech difficulty, altered mental status, ataxia, vision loss, and neglect determined by retrospective chart review), and symptom duration (<180, 180-270, and >270 min). Variables were then included in a multivariable model. Four hundred (56.4%) had a cerebrovascular event (TIA, infarct, or hemorrhage). Of the cases that the EMS crews identified as stroke, 72% were stroke or stroke mimic. EMS crews correctly identified 230 of these 400 patients as suspected stroke, yielding a sensitivity of 57.5% and PPV of 69.1%. Using both dispatcher and crew impression, 320 were correctly identified, which increased sensitivity to 80% and reduced PPV to 50.9%. Of the 400 cases with a confirmed vascular event, 87 (22%) were identified by the EMS crew alone, 90 (23%) by dispatcher alone, and 143 (36%) were identified by both (Figure 1). Of the 126 acute stroke patients with symptom onset < 6 h and NIHSS ≥ 5, EMS crews identified 101 (80.2%). Dispatchers correctly identified an additional 13 (10.3%) of these cases (Figure 1).
In univariate analysis, hemorrhagic stroke, NIHSS, and symptom duration were associated with a correct diagnosis by   Table 1). NIHSS, shorter symptom duration, and weakness were independently associated with a correct diagnosis in the multivariable model ( Table 2).
Comparing subjects with a vascular event who were identified by dispatchers alone to all other subjects in a univariate analysis, dispatchers alone were more likely to identify patients with TIA, low NIHSS, and vision loss at presentation; dispatchers were less likely to identify patients with altered mental status at presentation ( Table 3). In a multivariable model, age, NIHSS, and vision loss were independently associated with identification by dispatchers but not EMS crews ( Table 4).

DiscussiOn
Intravenous rt-PA and ET have been shown to dramatically improve outcomes in eligible subjects. ET is not widely available, and transfer times between hospitals are often long (10,11), so it is critical to identify stroke patients in the field and triage them to appropriate hospitals based on both diagnosis and suspected presence of a large vessel occlusion. Although scales exist to identify large vessel occlusion (12), EMS must first recognize that the patient is having a stroke.
The sensitivity of EMS crew identification of stroke was low at 57.5%. Philadelphia adheres to the Pennsylvania Statewide Advanced Life Support Protocols and Basic Life Support Protocols, which include the Cincinnati prehospital stroke scale (CPSS) for evaluation of suspected stroke patients (http://pehsc. org/wp-content/uploads/2015/06/Statewide_ALS_Protocols-2015-FINAL-06-01-15.pdf, http://pehsc.org/wp-content/uploads/ 2014/05/Statewide_BLS_Protocols_Final_020915.pdf ). Although our results are lower than the CPSS sensitivity reported in some studies (13,14), our results are similar to other studies in large metropolitan areas (5)(6)(7)(8)(9). This highlights the variability in EMS performance across geographic areas. Interestingly, we found that dispatchers identified 90 stroke patients that EMS crews did not. However, the collective dispatcher and EMS crew sensitivity exceeded that of previously reported EMS sensitivity. This builds on prior data, which showed dispatchers had greater diagnostic sensitivity than EMS crews (15). Civilians calling to report a suspected stroke are often correct in their initial impression (16), and it is possible that they raise the dispatcher's suspicion for stroke. EMS dispatchers in Philadelphia perform a verbal FAST screen (face, arm, speech, time) in accordance with their proprietary emergency medical dispatching software. Differences in compliance with prehospital stroke scales among dispatchers and crews or changes in symptoms over time could also account for these findings. Our finding that patients who were identified by dispatchers alone were more likely to have TIA, low NIHSS, and vision problems suggest that subtle and/or improving symptoms may be a factor in the discrepancy between EMS crew and dispatcher diagnosis. However, a combined approach to identification of stroke, which identified ischemic stroke that was identified by either the dispatcher or the EMS crew, was the most sensitive for identification of strokes with symptom duration <6 h and NIHSS ≥5. Given this, it seems that incorporating dispatcher impression into EMS prenotification protocols may be a viable option to improve prehospital recognition of stroke and maximize the use of rt-PA and ET. On the other hand, our data also illustrate the trade-offs of such an approach. Using EMS crew or dispatcher impression increased sensitivity, but PPV dropped from 69.1 to 50.9%. This suggests that a protocol incorporating both crew and dispatcher impression would increase the proportion of patients without stroke who were incorrectly triaged to a stroke center, which could be detrimental, particularly if local stroke centers are already operating near maximum capacity.
The retrospective nature of our single-center chart review has limitations. First, although Pennsylvania statewide EMS protocols include the CPSS for evaluation of suspected stroke patients, the database does not record whether this scale was used. As a result, we could not analyze how the use of a prehospital stroke scale affected EMS diagnostic accuracy. Second, HUP was a certified stroke center during the study period, so, EMS crews may have triaged more definitive, severe strokes to our center. If true, we may have overestimated EMS diagnostic accuracy. A third limitation was that we could not distinguish whether an ALS or BLS ambulance was dispatched on a particular call. While the paramedics staffing ALS ambulances function full-time in that role, during the study period, BLS ambulances were staffed with firefighter-EMTs who would work in an ambulance for several shifts, then rotate back to fire apparatus. As a consequence, their diagnostic accuracy may not have been comparable to that of paramedics. Finally, we also assumed that the database input fields truly reflect dispatcher and crew clinical impression. Given the time pressures facing EMS, recorded diagnostic impressions may not be an accurate assessment of their clinical suspicion and thus underestimate their diagnostic sensitivity.
Variability in prehospital identification of stroke across the US may limit the effectiveness of regionalized systems of care. Focused training initiatives are needed to improve EMS recognition of stroke and to ensure the use of prehospital stroke scales. Because dispatchers in this study identified stroke patients that crews did not, incorporating dispatcher diagnosis into prehospital interventions, EMS prenotification, or prehospital telemedicine, may increase the impact of these interventions.

eThics sTaTeMenT
The study protocol was approved by the Hospital of the University of Pennsylvania Institutional Review Board. auThOr cOnTriBuTiOns JJ: protocol development, data abstraction, data analysis, data interpretation, manuscript drafting/revising. RB: protocol development, data interpretation, manuscript drafting/revising. MA: data abstraction, data analysis, manuscript drafting/revising. WP: data analysis, manuscript drafting/revising. MG: data abstraction, manuscript drafting/revising. GD: protocol development, data interpretation, manuscript drafting/revising. CM and SM: data interpretation, manuscript drafting/revising. BC: protocol development, data interpretation, manuscript drafting/revising. MM: protocol development, data abstraction, data analysis, data interpretation, manuscript drafting/revising.

FunDing
Funded by a pilot grant from the Leonard Davis Institute of Health Economics at the University of Pennsylvania.