Dispatcher Recognition of Stroke Using the National Academy Medical Priority Dispatch System
Background and Purpose— Emergency medical dispatchers play an important role in optimizing stroke care if they are able to accurately identify calls regarding acute cerebrovascular disease. This study was undertaken to assess the diagnostic accuracy of the current national protocol guiding dispatcher questioning of 911 callers to identify stroke (QA Guide version 11.1 of the National Academy Medical Priority Dispatch System).
Methods— We identified all Los Angeles Fire Department paramedic transports of patients to University of California Los Angeles Medical Center during the 12-month period from January to December 2005 in a prospectively maintained database. Dispatcher-assigned Medical Priority Dispatch System codes for each of these patient transports were abstracted from the paramedic run sheets and compared to final hospital discharge diagnosis.
Results— Among 3474 transported patients, 96 (2.8%) had a final diagnosis of stroke or transient ischemic attack. Dispatchers assigned a code of potential stroke to 44.8% of patients with a final discharge diagnosis of stroke or TIA. Dispatcher identification of stroke showed a sensitivity of 0.41, specificity of 0.96, positive predictive value of 0.45, and negative predictive value of 0.95.
Conclusions— Dispatcher recognition of stroke calls using the widely employed Medical Priority Dispatch System algorithm is suboptimal, with failure to identify more than half of stroke patients as likely stroke. Revisions to the current national dispatcher structured interview and symptom identification algorithm for stroke may facilitate more accurate recognition of stroke by emergency medical dispatchers.
Emergency medical dispatchers play an important role in optimizing acute stroke care and facilitating rapid transport by initiating pre-arrival instructions to callers and dispatching emergency resources at the appropriate high level of priority. Studies in the United States in the 1990s, however, demonstrated dispatchers were only able to correctly identify approximately one-third to one-half of patients with stroke or TIA eventually diagnosed.1,2 A more recent study of emergency medical service calls in Germany similarly found that dispatchers recognized stroke in only 51% of stroke patients.3
In response to the availability of thrombolytic stroke therapy, a new algorithm for dispatcher use in interrogating 911 callers to identify stroke was implemented nationally in 2000. The Quality Assurance (QA) Guide version 11.1 of the Medical Priority Dispatch System (MPDS, Priority Dispatch Corporation) is the most widely used dispatcher guide in the United States. This version of the MPDS consists of interview protocols to identify 33 conditions, with stroke constituting condition 28 (Table 1). When a stroke is identified, the QA Guide instructs dispatchers that “Some strokes can now be effectively treated, but the time for successful therapy is quite short. Lights-and-sirens are not recommended; however, there should be a sense of urgency. Stroke must receive an immediate response that is not subject to delay.”
The separate interview protocol for stroke and high prioritization of stroke in the QA Guide version 11 represent progress over previous instruction manuals for dispatchers, which sometimes failed to recognize stroke as a separate entity or as an emergency requiring high-priority response. However, the stroke interview algorithm used has not been prospectively evaluated.
The goal of this study was to examine the ability of emergency medical dispatchers using MPDS algorithms to recognize stroke in a cohort of all consecutively transported patients in an urban emergency medical service region.
Subjects and Methods
All patients transported by Los Angeles Fire Department ambulances to the University of California Los Angeles Medical Center Emergency Department during a 12-month period from January 1, 2005 to December 31, 2005 were identified in a prospectively maintained database, the Los Angeles emergency medical service agency Trauma and Emergency Management Information System. For Trauma and Emergency Management Information System, paramedics complete 25 data fields on every patient they transport. Among these patients, neurological symptom patients were identified by analysis of the Trauma and Emergency Management Information System chief symptom field. In this field, paramedics categorize the patient’s chief symptom among 29 response options. Previous studies have demonstrated that 7 of the 29 symptom response options reflect neurological processes and identify nearly all stroke patients: (1) altered level of consciousness; (2) local neurological signs; (3) seizure; (4) syncope; (5) head pain; (6) nausea/vomiting; and (6) the cluster category of weak/dizzy.4 Examples of the 22 categories that are not directly neurologically relevant include chest pain, allergic reaction, abdominal pain, and shortness of breath.
The hospital administrative database for the same 12-month period was queried for field identifying the final International Statistical Classification of Diseases and Related Health Problems coded discharge diagnosis for all patients admitted to the hospital. All patients with a final International Statistical Classification of Diseases and Related Health Problems 9 discharge diagnosis consistent with ischemic/hemorrhagic stroke or TIA (codes 430–437) during the study period were identified.
For all patients presenting with neurological symptoms to paramedic personnel and all patients with a final discharge diagnosis of acute cerebrovascular disease, the prehospital medical record (paramedic “run sheet”) was examined to abstract the MPDS code assigned by the dispatchers. For paramedic run sheets with missing or illegible dispatch codes, incident numbers were used to query a database maintained by the Los Angeles Fire Department to recover as many dispatch codes as possible. Stroke patients were admitted to a specialized stroke neurology service at a tertiary university-affiliated medical center. Investigations typically included multimodal neuroimaging, including diffusion-weighted MRI, which is among the most sensitive means to diagnose ischemic stroke. The study was approved by the University of California Los Angeles Institutional Review Board.
Test performance characteristics were analyzed based on the directly measured number of true-positives, false-positives, and false-negatives. The number of true-negatives was imputed using the equation: true-negatives=number of patients with MPDS code retrieved/rate of neurological symptoms−(true-positives+false-positives+false-negatives), where rate of neurological symptoms is the rate of neurological symptoms in the transported population. Sensitivity, specificity, predictive values, and likelihood ratios for the dispatcher assigned stroke MPDS codes were calculated in standard fashion, with the final discharge diagnosis as the reference standard. Confidence intervals for these values were calculated using efficient-score method.5
During the 12-month study period, 3474 patients were transported by Los Angeles Fire Department ambulances to University of California Los Angeles Medical Center. Among these, 1283 (36.9%) involved patients with chief symptoms potentially referable to the nervous system and potentially related to acute cerebrovascular disease; 96 patients had a final diagnosis of stroke or TIA (2.8% of total transports). Of the patients identified with neurological chief symptoms, 871 (67.9%) had retrievable dispatcher-assigned MPDS codes. In the remainder, MPDS codes were not recorded or illegible on the paramedic run sheets and could not be retrieved from the Los Angeles Fire Department mainframe because of missing or invalid incident numbers.
Among the 871 patients, there were 58 patients with a MPDS dispatch code of stroke, of whom 26 (44.8%) had a final diagnosis of acute cerebrovascular disease (true-positives) and 32 (55.2%) had a nonstroke diagnosis (false-positives). The leading final discharge diagnoses for the 32 false-positive patients were: cardiac/respiratory related (19%), vertigo/syncope/altered level of consciousness (22%), hypotension/hypovolemia (13%), malignancy (9%), and infection (9%).
Eight-hundred thirteen patients had nonstroke MPDS codes and within this group there were 38 (4.7%) with stroke discharge diagnoses (false-negatives) and 775 (95.3%) with nonstroke discharge diagnoses (true-negatives). For the group of 38 stroke patients not correctly identified by dispatchers, the most commonly (>5% frequency) assigned nonstroke MPDS codes were not alert (29.6%), unconscious (16.1%), and cardiac (10.7%). Test performance characteristics for the MPDS dispatch codes for stroke are shown in Table 2. Although specificity was high, sensitivity and positive predictive value were modest.
The mean (±SD) age for stroke patients correctly and incorrectly identified by dispatchers did not differ (correct 75.5±19.7 vs incorrectly 71.9±21.4). With regard to gender, 71.3% of male stroke patients were misidentified compared to 56.2% of female stroke patients; however, this difference failed to achieve significance (P=0.187).
In this study, the current national algorithm for emergency medical dispatchers diagnosis of stroke demonstrated only modest sensitivity and positive predictive value. Over the 12-month study period, emergency medical service dispatchers correctly recognized 45% of patients with a final discharge diagnosis of stroke or TIA as having acute cerebrovascular disease. Conversely, more than half of the patients assigned a stroke code by emergency medical dispatchers were nonstroke patients. Our findings confirm and expand those of a study of MPDS dispatch algorithm in the city of San Diego published after the initial submission of this manuscript.16 Both studies found that the MPDS exhibited only modest positive predictive value, 45% in Los Angeles and 42.5% in San Diego, indicating that the majority of patients identified as having stroke by dispatchers using the MPDS do not actually have stroke. As we identified stroke final diagnoses in all transports, not just those with dispatcher diagnosis of stroke, we were able to delineate specificity and negative predictive value and more accurately delineate sensitivity than in the San Diego investigation. We found that high rates of specificity and negative predictive value, but only modest sensitivity performance. More than half of patients who had a final diagnosis of stroke were not recognized as having stroke by dispatchers.
A content analysis of the MPDS stroke recognition algorithm suggests inadequate emphasis on motor stroke symptoms as a likely cause of suboptimal performance. Dispatchers are instructed to ask directly only about symptoms of talking abnormally and decreased level of consciousness. These symptoms and signs are present in only ≈50% to 65% of strokes. Motor weakness, especially asymmetrical, is the most discriminating sign of stroke in the prehospital setting. Motor symptoms are present in 80% to 90% of all stroke patients, and an even greater proportion of patients for whom the 911 call system is activated.4,6–10 However, inquiry regarding motor symptoms only occurs in the QA Guide version 11 protocol if the caller first states the patient is having a stroke.
In a study by Italian investigators11 of dispatcher phone encounters with 177 consecutive potential stroke patients, among 8 questions evaluated, only 3 were found to be statistically associated with stroke: mouth asymmetry, arm weakness, and leg weakness. Demonstrating poor sensitivity and specificity for stroke were queries regarding level of consciousness, comprehension, speech output, headache, and vertigo. However, level of consciousness and speech abnormality are emphasized in the current MPDS algorithm, despite their lack of sensitivity and specificity for stroke. The misidentification of multiple stroke-mimicking conditions as representing stroke that we documented, including multiple cardiovascular and respiratory conditions, is to be expected given the many other conditions that disturb language output and alertness and the large number of true strokes that leave speech output and alertness undisturbed.11
Contributing to poor dispatcher recognition of stoke is the lack of public awareness of stroke symptoms. In a population-based survey conducted in 1995, a majority of elderly individuals failed to list at least 1 stroke warning sign.12 In studies of emergency medical service calls related to stroke, “stroke” was identified as the problem by the caller less than half of the time (range, 19.8%–44%),3,13 and even when patients use the word “stroke” they are frequently assigned nonstroke codes.2,14
This study has some limitations, many inherent to research conducted using prehospital forms and administrative databases. Data from 32% of the paramedic run sheets had MPDS codes that were either illegible or not recorded, a rate typical in prehospital studies.15 We could not identify any systematic difference between cases in which a transport form was retrievable and interpretable and cases in which forms were not available; however, it not possible to fully exclude a selection bias. Because of the unavailable forms, the specificity and negative predictive values derived in this study were arrived at by combining direct measures with the assumption of no difference between available and unavailable form cases. However, for low-frequency conditions, test utility is better indicated by positive predictive value and sensitivity than by negative predictive value and specificity. Sensitivity and positive predictive value were directly measured in this study.
This study shows that the current national algorithm for emergency medical service dispatcher recognition of stroke is suboptimal, with emergency medical dispatchers failing to recognize stroke in more than half of true stroke 911 calls. Revision of the algorithm to emphasize asymmetrical motor deficits and de-emphasize altered level of consciousness likely would improve dispatcher performance without increasing interview duration and merits prospective study. Furthermore, the results emphasize that paramedics and emergency medical technicians, as the next stage of contact in the chain of prehospital care, must perform well in identifying stroke because dispatch codes will often be incorrect.4,17 Accurate identification of acute stroke patients in the field permits pre-arrival notification of the receiving hospital, clearing of the CT or MR scanner for rapid imaging, earlier assessment by stroke team physicians, more frequent recanalization therapy treatment,18 and direct routing of appropriate patients to designated stroke centers.
Additionally, efforts to educate the public on the symptoms of stroke should continue because this may enhance the ability of callers to communicate relevant information and assist dispatchers in identifying stroke.19 Finally, public education regarding the appropriate use of 911 for stroke should continue as the emergency medical services system remains the timeliest means of accessing acute stroke care.20
Sources of Funding
Supported by NIH-NINDS Award P50 NS044378 and Heart and Stroke Foundation of Canada Fellowship Award (B.H.B.). Preliminary results from this study were presented at the 2007 International Stroke Conference, San Francisco, Calif.
- Received December 17, 2008.
- Accepted January 6, 2009.
Kothari R, Barsan W, Brott T, Broderick J, Ashbrock S. Frequency and accuracy of prehospital diagnosis of acute stroke. Stroke. 1995; 26: 937–941.
Handschu R, Poppe R, Rauss J, Neundorfer B, Erbguth F. Emergency calls in acute stroke. Stroke. 2003; 34: 1005–1009.
Kidwell CS, Starkman S, Eckstein M, Weems K, Saver JL. Identifying stroke in the field: Prospective validation of the Los Angeles Prehospital Stroke Screen (LAPSS). Stroke. 2000; 31: 71–76.
Herman B, Leyten AC, van Luijk JH, Frenken CW, Op de Coul AA, Schulte BP. Epidemiology of stroke in Tilburg, the Netherlands. The population-based stroke incidence register: 2. Incidence, initial clinical picture and medical care, and three-week case fatality. Stroke. 1982; 13: 629–634.
Wade DT, Wood VA, Hewer RL. Recovery after stroke—the first 3 months. J Neurol Neurosurg Psychiatry. 1985; 48: 7–13.
Bogousslavsky J, Van Melle G, Regli F. The Lausanne stroke registry: Analysis of 1,000 consecutive patients with first stroke. Stroke. 1988; 19: 1083–1092.
Libman RB, Sacco RL, Shi T, Tatemichi TK, Mohr JP. Neurologic improvement in pure motor hemiparesis: Implications for clinical trials. Neurology. 1992; 42: 1713–1716.
Camerlingo M, Casto L, Censori B, Ferraro B, Gazzaniga G, Partziguian T, Signore M, Panagia C, Fascendini A, Cesana BM, Mamoli A. Experience with a questionnaire administered by emergency medical service for pre-hospital identification of patients with acute stroke. Neurol Sci. 2001; 22: 357–361.
Mosley I, Nicol M, Donnan G, Patrick I, Dewey H. Stroke symptoms and the decision to call for an ambulance. Stroke. 2007; 38: 361–366.
de la Ossa NP, Sanchez-Ojanguren J, Palomeras E, Millan M, Arenillas JF, Dorado L, Guerrero C, Abilleira S, Davalos A. Influence of the stroke code activation source on the outcome of acute ischemic stroke patients. Neurology. 2008; 70: 1238–1243.
Schwamm LH, Pancioli A, Acker JE III, Goldstein LB, Zorowitz RD, Shephard TJ, Moyer P, Gorman M, Johnston SC, Duncan PW, Gorelick P, Frank J, Stranne SK, Smith R, Federspiel W, Horton KB, Magnis E, Adams RJ. Recommendations for the establishment of stroke systems of care: Recommendations from the Am Stroke Association’s task force on the development of stroke systems. Circulation. 2005; 111: 1078–1091.