Background and Purpose: As many as half the patients presenting with acute stroke access medical care through the ambulance service. In order to identify and triage these patients effectively as life-threatening emergencies, telephone-based ambulance software must have high sensitivity and specificity when using verbal descriptions to identify such patients. Software-based clinical coding was compared with the patient’s final clinical diagnosis for all patients admitted by ambulance to North Hampshire Hospital (NHH) emergency department (ED) over a 6-month period to establish the ability of telephone-based triage to identify patients with likely stroke accurately.
Methods: All emergency calls to South Central Ambulance Service over a 6-month period resulting in a patient being taken to NHH ED were reviewed. The classification allocated to the patient by ambulance advanced medical priority dispatch software (AMPDS version 11.1) was compared with the final clinical diagnosis made by a doctor in the ED.
Results: 4810 patients were admitted to NHH during the study period. Of these, 126 patients were subsequently diagnosed as having had a stroke. The sensitivity of AMPDS software for detecting stroke in this sample was 47.62%, specificity was 98.68%, positive predictive value was 0.49 and negative predictive value was 0.986.
Conclusions: Fewer than half of all patients with acute stroke were identified using telephone triage on the initial emergency call to the ambulance service. Less than one quarter received the highest priority of ambulance response. This first link in the chain of survival needs strengthening in order to provide prompt and timely emergency care for these patients.
Statistics from Altmetric.com
Stroke is one of the most significant of all disease pathologies, both in terms of the devastating impact on quality of life and the cost to the health economy.
The estimated 110 000 cases of stroke in the UK each year result in the disease being the third most common cause of death in the UK and the single most common cause of severe disability, with more than 250 000 people living with stroke-related disabilities.1 The past decade has seen a recognition of the need to treat stroke as an acute emergency,2–4 with studies showing that prompt treatment and thrombolysis when appropriate can significantly reduce mortality and morbidity. Department of Health figures estimate that treating stroke as an emergency would result in more than 1000 stroke patients a year regaining independence rather than dying or being left dependent on others as a result of significant disability.5 The Department of Health is now driving forward a national programme to improve the care given to patients with an acute stroke in order to reduce the socioeconomic impact of this disease.1 6
Approximately 20–47% of patients with an acute stroke in the UK access medical care through the ambulance service rather than through primary care.7–9 The telephone consultation used by the ambulance service is now a common initial point of access for healthcare, and presents unique challenges compared with traditional face-to-face consultation and diagnosis. The UK ambulance service receives its emergency workload through telephone consultation, and over the past decade has progressed from an unreliable subjective assessment of the emergency call to a more objective assessment using a structured telephone protocol to establish a chief complaint and allocate a predetermined response priority. Advanced medical priority dispatch software (AMPDS), now used by many European and North American ambulance services, contains stroke-related questions specifically questioning the caller about speech or movement problems, numbness or tingling and past stroke history. Patients subsequently categorised to a chief complaint of ‘stroke’ are graded as a life-threatening (category A) emergency and allocated an 8-minute ambulance response only if they are unconscious; otherwise they are classified to an urgent (category B) 19-minute response.
The ability of telephone-based pathways to identify patients with acute stroke accurately is key to the efficient allocation of limited ambulance resources and a prompt response to this time-critical emergency. Little work has been carried out examining the ability of telephone triage to identify specific life-threatening conditions correctly. Previous work from this ambulance service has shown that telephone triage is limited in its ability to identify acute coronary events correctly,10 but the ability of the same software to identify patients with an acute stroke correctly is unknown. We therefore carried out a retrospective study comparing the ‘chief complaint’ code allocated by the AMPDS telephone triage software with the final diagnosis made by a doctor at a hospital emergency department (ED) in order to establish the ability of telephone-based consultation to identify acute stroke and provide a strong link in the stroke ‘chain of survival’.
Box 1 Statistical definitions
Sensitivity: Chance of having a positive test in patients with the test condition
Specificity: Chance of having a negative test in patients without the condition
Positive predictive value: Chance of having test condition given a positive test result
Negative predictive value: Chance of not having test condition given a negative test result
Prevalence: True positives plus false negatives for the test condition as a proportion of the population
MATERIALS AND METHODS
All emergency calls to the Hampshire Division of South Central Ambulance Service are received by a central control room where data relating to the calls are logged and stored on a mainframe computer. All emergency (999) calls are answered by a call taker who uses AMPDS software (version 11.1; Priority Dispatch Corp, USA) to triage the call. A series of structured questions is used to determine the patient’s ‘chief complaint’ from the information provided by the caller, according to an algorithm. This then allows the call taker to ask more condition-specific predetermined questions in order to identify the clinical condition of the patient. From this information a code is allocated to the call, which has a predetermined resource/priority level assigned to it.
All hospital notes for patients transported by the South Central Ambulance Service to North Hampshire Hospital ED between 1 December 2006 and 31 May 2007 were reviewed and those identified with a final ED diagnosis of acute stroke or transient ischaemic attack (TIA) were identified. In practice, no distinction was made between acute stroke and TIA because they are essentially the same pathology and, in terms of ambulance response priority, are identical. Corresponding ambulance records for these patients were examined to document the initial emergency call ‘chief complaint’ and subsequent priority given for the ambulance response.
Data were analysed for sensitivity and specificity using a web-based statistical software calculator (http://www.hutchon.net/EPRval.htm). Data were classified as follows: true positive—patients classified by AMPDS as having an acute stroke, with a confirmed acute stroke diagnosis at hospital; true negative—patients classified to an AMPDS criteria other than acute stroke, who had a diagnosis other than acute stroke at hospital; false positive—patients classified by AMPDS as having an acute stroke, who had a diagnosis other than acute stroke at hospital; false negative—patients classified to an AMPDS criteria other than acute stroke, with a confirmed acute stroke diagnosis at hospital.
For statistical definitions see box 1.
In the 6-month period between 1 December 2006 and 31 May 2007, 4810 patients were admitted to North Hampshire Hospital ED. A total of 126 patients was subsequently given a discharge diagnosis by the ED as acute stroke.
The mean age of confirmed acute stroke patients was 74.3 years, with a female : male ratio of 1.1 : 1.0. Of the 126 patients with confirmed acute strokes, 60 had been allocated a ‘chief complaint’ of ‘stroke’ by the AMPDS software, but 66 patients were allocated to a ‘chief complaint’ other than ‘stroke’. A total of 62 patients was classified by AMPDS software to the category of acute stroke/TIA, but was subsequently found not to have had a stroke following clinical examination in hospital.
Allocation to a category of ‘stroke’ by AMPDS-based telephone triage was as follows: true positive 60; false negative 66; true negative 4622; false positive 62. Statistical analysis gives the following results: specificity 98.7%; sensitivity 47.6%; prevalence 2.6%; positive predictive value 49.2% (95% CI 40.5 to 57.9); negative predictive value 98.6% (95% CI 98.2 to 98.9).
Of the 66 patients with a hospital diagnosis of acute stroke who were previously categorised to a non-stroke code by AMPDS software (false negatives), AMPDS classifications were made as shown in table 1.
Of the patients with confirmed strokes, ambulance responses were allocated as shown in table 2 (n = 103 for available data).
The prompt and early treatment of acute stroke/TIA has the ability to limit morbidity and mortality and have a significant socioeconomic impact. The past decade has seen recognition of the need to treat this condition as a medical emergency rather than a futile condition, with this now recognised as a priority in healthcare delivery both in Europe and the USA. In order to achieve this, a series of integrated key stages, as outlined in the American Stroke Association’s ‘stroke chain of survival’, need to be functioning in order to deliver appropriate and timely treatment.11 Ambulance involvement is key in the early stages by promoting patient awareness, correctly identifying the symptoms presented by the emergency call, attending the patient promptly and pre-alerting the receiving hospital. A large proportion of these patients will initially present to the ambulance service through a telephone-initiated call. Accurate diagnosis by verbal communication only, often through a third party, is particularly challenging and is compounded by the often limited clinical knowledge of the ambulance call taker. The American Heart Association/American Stroke Association Expert Panel on Emergency Medical Services Systems12 and current American Heart Association resuscitation guidelines13 have identified the importance of the accurate identification of these patients at first telephone contact and have recommended that priority is given to identifying acute stroke patients rapidly and accurately by ensuring that emergency medical service (EMS) communicators recognise stroke signs and symptoms as reported by callers. Although educational programmes for call takers can improve the diagnostic accuracy for acute stroke,14 many ambulance services use call takers with no specific medical training and therefore rely on predetermined structured written questions. In the UK, most call takers have no medical training and in order to provide a more consistent, accurate and effective telephone-based identification of these patients, all ambulance services in the UK since 2002 now use an algorithm-based clinical decision tool that attempts to improve the accuracy of telephone triage by using a series of specific structured questions to ascertain the nature of the problem. In the UK, most ambulance trusts use AMPDS software15 as their triage tool. Having allocated the patient to a specific clinical category, the patient is then allocated to a response priority according to predetermined categorisation by the Department of Health. Patients identified as a likely stroke or TIA are allocated to the highest level of emergency (category A—life threatening), requiring an ambulance attendance to 75% of calls within 8 minutes only if the caller states that the patient is unconscious. Simple stroke/TIA and other diagnoses considered less serious are allocated either to category B (urgent) requiring ambulance attendance at 95% of calls within 19 minutes. Category C (non-urgent) classification requires ambulance attendance at 100% of calls within 60 minutes and is reserved for what are considered minor ailments.
This study has demonstrated the challenges of using telephone consultation to identify patients likely to have had an acute stroke or TIA accurately. Of all patients finally diagnosed with acute stroke/TIA, AMPDS software only allocated a correct classification in approximately half of these patients. A significant number of patients with acute stroke was allocated to a category of ‘collapse/unconscious’. Fortuitously, most of these complaints were allocated a category A response. This has been reported in other studies in which alternative symptoms are considered sufficiently serious still to generate a high priority response.16 Because the Department of Health only requires acute stroke to receive a category B (19-minute) response, only 22.3% patients with acute stroke/TIA received the highest priority of response. Of the remaining patients in this study, 74.8% received a category B 19-minute response, but a small proportion of patients (2.9%) received a category C 60-minute response. Ultimately the ambulance prioritisation is more important for the patient than the actual categorisation made by the AMPDS software. The prioritisation is, however, determined by the Department of Health, based on the initial AMPDS coding. A review of the prioritisation of AMPDS codes may be necessary based on this study if more stroke patients are to be allocated an appropriate response.
Previous studies have showed that EMS call takers using their own subjective assessment correctly identified 31%16 17 to 52%8 of patients with acute stroke when compared with the final diagnosis. Even when patients or relatives specifically mentioned the word ‘stroke’, the call taker would often conclude with an alternative diagnosis.17 Inaccuracy may be due to the brief interactions that the EMS personnel have with patients, but the 15–21 h of dispatcher training with no specified training in neurological diseases was also considered to be a major factor.8 An analysis of predictive words used by patients when calling an EMS dispatcher showed that the word ‘stroke’ was highly predictive of an acute stroke (positive likelihood ratio 2.27) and that the presence of at least one of four words/phrases (stroke, facial droop, weakness/fall, or impaired communication) predicted 80% of all stroke calls.18 These key features are incorporated into the AMPDS questions asked of the caller, but have not managed to achieve such diagnostic accuracy.
Further studies are needed to refine the identification of acute stroke by telephone consultation alone. Structured questioning is likely to be more consistently effective than allowing call takers the freedom to ask questions subjectively, particularly if they have limited medical knowledge and are interrogating callers under stressful conditions, as is often the case. Patients who have collapsed or are unconscious will receive a high priority ambulance response, but may nevertheless have an underlying stroke as the cause of their demise. The current recommendations for a 19-minute response to acute stroke/TIA patients should be reviewed with the aim of upgrading to an 8-minute response. Strengthening this key stage in the stroke chain of survival is important if a prompt ambulance response is to be delivered to these patients.
Contributors: CDD, FT and MA were involved in the study protocol. MA, PK and CDD were responsible for clinical data collection. All were responsible for data analysis and writing up of the results. All made critical revisions to the manuscript. CDD is the guarantor.
Competing interests: None.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.