Telestroke consultation can accurately diagnose ischemic stroke mimics

Abstract

Introduction

Telestroke (TS) networks are standard in many areas of the US. Despite TS systems having approximately 33% mimic rates, it is unknown if TS can accurately diagnose patients with acute ischemic stroke (AIS) versus stroke mimics.

Methods

We performed a retrospective review of consecutive TS consults to 27 TS sites in six states during 2018. Clinical information and diagnosis were extracted from discharge records and compared to those from the TS consult. Discharge diagnoses were verified and coded into 12 categories. Cases without a clear discharge diagnosis and intracerebral haemorrhage were excluded. We report agreement and a Cohen’s kappa between TS and discharge diagnoses for the category of AIS/transient ischemic attack (TIA) versus stroke mimic.

Results

We included 404 cases in the analysis (mean age 66 years; 54% women). Of these, 225 had a TS diagnosis of AIS/TIA; 102 (45%) received intravenous tissue plasminogen activator. Our study demonstrated a high diagnostic agreement for AIS/TIA (88%) with a kappa of 0.75 for stroke and mimics. Of the 179 patients diagnosed with a stroke mimic on TS, 27 (15%) were diagnosed with AIS/TA by discharge. TS mimic diagnosis had a positive predictive value (PPV) of 85% and a negative predictive value (NPV) of 90%; TS diagnosis of stroke/TIA had PPV 90%, NPV 85%.

Discussion

We found excellent correlation between TS and discharge diagnoses for patients with both stroke and stroke mimics. This suggests that TS systems can accurately assess a wider variety of patients with acute neurologic syndromes other than AIS.

Keywords

Teleneurology telemedicine teleconsulting quality measures diagnostic accuracy,stroke

Introduction

Telestroke (TS) was first coined by Levine in 1999 as a system with the purpose of maximizing the number of stroke patients given effective therapies and increasing the adherence to stroke protocols through telecommunication counseling.¹ Since that time, TS utilization has become highly prevalent throughout not only the US, but also throughout the world.²,³ It is particularly utilized in rural areas, where community hospitals often lack the necessary expertise to effectively provide neurological services.⁴ Subsequent studies demonstrated that TS is accurate for administration of the National Institute of Stroke Scale (NIHSS) and increases adherence to stroke protocols.²,³,⁵,⁶

In multiple networks, patients with a stroke mimic consist of more than a third of the TS consultations.^7–9 However, despite this high prevalence, the patients with stroke mimics are typically excluded from quality studies of TS networks.¹⁰ In particular, the accuracy of TS networks in diagnosing stroke mimics is poorly understood. Therefore, we aimed to determine the accuracy of neurologic diagnosis in patients evaluated via TS consultation in a large hub-and-spoke network.

Methods

We performed a single-centre, retrospective review of the University of Utah’s TS service.

Network characteristics

Our network is a hub–spoke model and, at the time of the study, consisted of 27 spoke sites across six states in the Intermountain West region. In 2018, TS consultation were provided by seven stroke-trained neurologists and two neuro-intensivists. While there are no restrictions to TS consultation, we encourage our spoke site providers to consult our TS service primarily for patients potentially eligible for revascularization therapies (either intravenous tissue plasminogen activator (tPA) or endovascular treatment). Thus, patients with subacute stroke and no need for acute management decision-making would typically by managed by phone-only consult, and not included in this study. Head computed tomography (CT) without contrast is typically performed at spoke sites prior to or concurrent with TS consultation (and in this case series, was done so in all patients but one). We are not typically consulted on patients with intracerebral haemorrhage (ICH), except when consulted prior to completion of the head CT and the diagnosis is not yet known.

The TS consultant can choose to conduct the consult via phone only or two-way video. For this study, we only analysed consults that were video-enabled as this allowed performance of a neurologic exam. All TS consultations in this analysis were thus performed with two-way video capability, with the consultant confirming and expanding the history provided by the community hospital staff, and a neurologic examination performed by the consultant and assisted by the on-site nurse or physician. In our TS network, it is standard to calculate the NIHSS for all patients with additional neurologic exam done as deemed necessary by the consultant. The TS consultant personally reviewed any neuroimaging at the time of consult with final interpretation by the site’s contracted radiologist. Post-TS consultation, the decision to transfer to a higher level of care is made based on the individual patient’s needs and site’s capabilities.

Inclusion/exclusion criteria

We included all consecutive video-enabled TS consults from January to December 2018, excluding patients who were under 18 years of age at time of consultation. We also excluded all patients with ICH, as ICH is readily diagnosed by head CT, patients with ICH are not typically managed by our TS network, and thus patients with ICH are not the focus of this analysis.

Data collection

Patient data regarding demographics, diagnosis at time of TS consultation, and final diagnosis on discharge (from the emergency department or, if admitted, from the hospital) were extracted from original clinical records from the spoke and hub. Diagnoses were extracted from the assessment provided in the consultation note for the TS consult and from the discharge summary for final discharge diagnosis. When multiple TS diagnoses were listed without priority, the first was used as the diagnosis. We excluded cases where the diagnosis by either TS consultation or discharging physician was not clear by documentation. There was one case of ICH in our data set and it was excluded.

Diagnostic categories

We categorized the diagnoses from TS consultation as acute ischemic stroke (AIS) and transient ischemic attack (TIA), or ischemic stroke mimic. In the analysis, AIS and TIA were combined as we felt that they represent similar pathophysiology and cannot always be distinguished accurately in a single encounter early in the clinical course, particularly prior to obtaining a brain magnetic resonance imaging (MRI). Two physicians, a neurology fellow (JP) and a vascular neurologist (JJM), independently reviewed all diagnostic codes and resolved disagreements with a consensus final diagnosis. We defined AIS/TIA using the definitions described in Sacco et al.¹¹ We defined stroke mimic in this study as a patient presenting with acute neurologic deficits thought initially by the emergency department (ED) provider to possibly be AIS or TIA, but upon evaluation by a vascular neurologist, including head CT at a minimum, and MRI when indicated, was found to not be AIS or TIA.

We then coded all AIS mimics into 10 sub-categories based upon prevalence in the overall sample. These codes included: complex migraine, psychiatric event, seizure, other central nervous system (CNS) processes, toxic/metabolic aetiologies, other systemic illnesses, unspecified altered mental status (AMS), peripheral cranial nerve (CN) lesion, and peripheral nervous system (PNS)/muscular skeletal (MSK) lesion (Figure 1). Psychiatric events included conversion, anxiety, malingering, and catatonic events. Seizure included post-ictal paresis. Other CNS processes included transient global amnesia, reversible cerebral vasoconstriction syndrome, tumour oedema, subdural hematoma, and unmasking of prior CNS lesions. Toxic/metabolic aetiologies included non-CNS infections, sepsis, metabolic derangements, and polypharmacy. Other systemic illnesses included hypertensive emergency, gastrointestinal bleed, and chronic heart failure exacerbation. Unspecified AMS was coded when the discharging physician did not give an aetiology of AMS. Peripheral CN lesions included Bell’s Palsy, peripheral vertigo, and microvascular CN palsies. PNS/MSK lesions included radiculopathy, fractures, and lumbago.

Figure 1.

Study paradigm, 476 charts were reviewed in total, 72 met exclusion criteria, participants for group by TS diagnosis with 225 having a diagnosis of AIS/TIA and 179 having AIS mimic. AIS mimics were then sub-categorized into 10 categories: complex migraine, psychiatric events, seizure/Todd’s paresis, other CNS processes, toxic/metabolic aetiologies, other systemic illnesses, unspecified AMS, peripheral CN lesion, and PNS/MSK lesion.

Statistics

To determine the accuracy of the TS system, we compared diagnostic categories between the TS consultation and final diagnosis at discharge from the ED or, if admitted, from the hospital. We compared patients with TS diagnosis of AIS/TIA versus stroke mimic by age and NIHSS using two tailed t-tests and by sex using X² test. Cohen’s kappa was calculated for diagnostic agreements between TS diagnosis and final discharge diagnosis along with the positive predictive and negative predictive values (PPV and NPV) of TS diagnosis for AIS/TIA and stroke mimic.

We conducted stepwise logistic regression to determine predictors of patients initially diagnosed with stroke mimic and found to have a final diagnosis of TIA/AIS. To determine if our results were robust when excluding patients with transient symptoms, we performed a sensitivity analysis where we excluded patients with a final diagnosis of TIA. To do so, we fit a stepwise backwards selection logistic regression model, set to p < 0.05, to the outcome of disagreement between TS and final diagnosis of AIS/TIA. Data analysis was performed using Stata 16.0 (StataCorp, College Station, TX).

Ethics

The University of Utah Institutional Review Board approved this study with exemption from informed consent by the participants.

Results

We reviewed 476 consecutive TS cases. We excluded 72 for lack of discharge diagnosis in the documentation and excluded the one case of ICH, leaving 404 cases for analysis. There were no paediatric cases to exclude in this sample. We found that 55.5% (n = 225) of our total stroke consultations resulted in a TS diagnosis of AIS (n = 198) or TIA (n = 27). In the 27 patients with a TS diagnosis of TIA, 21 were eventually diagnosed with AIS as their final diagnosis. Of the 404 patients, mean age was 66.4 years old (interquartile range (IQR) 55–79 years old). Patients with a TS diagnosis of AIS/TIA had a mean age of 71.3 years (IQR 64–82), while patients with a TS diagnosis of stroke mimic were younger (mean age 60.3 years old, IQR 19–98, p < 0.001). The sex distribution of TS diagnosis of AIS/TIA versus stroke mimic was similar: AIS/TIA diagnoses were 51% female and stroke mimics were 57% female (p = not significant). Our sample was predominantly Caucasian. Stroke severity, as measured by the NIHSS, was higher in patients who received a TS diagnosis of AIS/TIA compared to stroke mimic (mean NIHSS score 6.2 versus 3.8, p < 0.001; Table 1).

Table 1.

Patient demographics.

	Total sample (N = 404)	TS diagnosis of AIS/TIA (n = 225)	TS diagnosis of AIS mimic (n = 179)	p-value
Age (years), mean (IQR)	66.4 (55–79)	71.25 (64–82)	60.4 (45–76)	p < 0.001
Female, n (%)	218 (53.8)	115 (51.1)	103 (57.2)	NS
Race/ethnicity, n (%)
Non-Hispanic white	377 (93.1)
Hispanic white	17 (4.2)
Native American	5 (1.2)
Asian	4 (1.0)
African American	1 (0.2)
Unknown	1 (0.2)
NIHSS score, mean (IQR)	5.13 (1–6)	6.2 (2–8)	3.8 (1–4)	p < 0.001
tPA administered	103	103	1
Transferred to hub hospital	85	73	12

TS: telestroke; AIS: acute ischemic stroke; TIA: transient ischemic attack; NS: not significant; NIHSS: National Institute of Stroke Scale; IQR: interquartile range; tPA: tissue plasminogen activator.

When comparing the TS diagnosis of AIS/TIA to discharge diagnosis, we find 89.8% concordance rate with the original diagnosis. Of those initially diagnosed with AIS/TIA on TS consultation, 23 (10.2%) had a stroke mimic diagnosed at discharge, the most common of which was complex migraine, other CNS processes, and non-stroke complications of systemic illnesses (Figure 2).

Figure 2.

Distribution of final diagnoses in patients initially diagnosed with AIS/TIA shows high percentage of diagnostic agreement. Of those initially diagnosed with AIS/TIA, 10.2% had a final diagnosis of AIS mimic. Most common, after an initial diagnosis of AIS/TIA, was complex migraine, other systemic illness, and other CNS processes. All percentages are expressed as a portion of the overall group (n = 225).

Of those initially diagnosed with stroke mimic, final discharge diagnoses were AIS/TIA in 15% (n = 27). In the stepwise logistic regression model fit to these 27 patients, we found that neither patient sex, race, ethnicity; TS provider; nor facility and door-to-call time were associated with AIS/TIA disagreement. However, patient age and NIHSS were associated. In the 27 patients with disagreement, the mean age was 72.2 years, while in those without disagreement (n = 152) it was 66.1 (p = 0.084); the mean NIHSS was 3.1 versus 5.3, respectively (p = 0.063). However, in the logistic regression model containing both terms, the pseudo R-squared was 0.05, suggesting that this model does not adequately explain the disagreement between TS and final diagnosis – that is, that age and NIHSS themselves cannot predict the transformation of TS mimic diagnosis to final diagnosis of AIS/TIAs. The remaining 85% of those initially diagnosed with stroke mimic maintained a non-AIS/TIA diagnosis at discharge. The most common discharge diagnosis among stroke mimics was complex migraine, which comprised 20% (n = 36) of those initially diagnosed with AIS/TIA, followed by psychiatric event (n = 22, 12%), and seizure (n = 19, 10%) (Figure 3).

Figure 3.

Distribution of final diagnoses for patients initially diagnosed with AIS mimic on tele-consultation. Of those initially diagnosed with AIS mimic, 15% were diagnosed with AIS/TIA at discharge, with an agreement of 85% for a diagnosis of AIS mimic. The most common final diagnosis was complex migraine, psychiatric events, and complications of seizure.

Diagnostic agreement between TS diagnosis and final discharge diagnosis as measured by the Cohen’s kappa was 0.75 for both AIS/TIA and stroke mimic groups. In our network, TS stroke mimic diagnosis had a PPV of 85% and an NPV of 90%. Similarly, a TS diagnosis of AIS/TIA conferred a PPV of 90% and an NPV of 85%. In the sensitivity analysis where we excluded 44 patients with a final diagnosis of TIA, the Cohen’s kappa was 0.79 and TS diagnosis of AIS/TIA conferred a PPV of 88% and an NPV of 92%.

In our population, 21% (n = 85) of the total sample were transferred from the spoke hospital to the hub hospital. Of AIS mimics, 6.7% (n = 12) were transferred. tPA was administered at the spoke hospital in 25.4% of the total consultations (n = 103). We had one case of suspected AIS mimic diagnosis by TS consult that was given tPA, and eight cases with discharge diagnosis of AIS mimic that were given tPA. Overall, 50.6% (n = 43) of patients receiving tPA were transferred to the hub hospital.

Discussion

We performed a single-centre, retrospective study to determine the diagnostic accuracy of TS consultation. We found a high diagnostic concordance between TS diagnosis of AIS/TIA or stroke mimic and final discharge diagnosis, including high PPVs and NPVs for TS diagnosis of either AIS/TIA or stroke mimic. Overall, this suggests that the TS system can accurately assess a wider variety of patients with acute neurological syndromes than only AIS/TIA. The ability to properly distinguish between AIS mimic and AIS/TIA has direct consequences to the patient’s health. Accurate assessments of AIS mimics can avoid unnecessary exposure to and cost of thrombolytics,¹² intensive care unit stay, laboratory work-up, and other charges related to inpatient admission. One study estimated the total of such excess costs of thrombolysis in a AIS mimic to be a median of approximately US$5400 per admission.¹³ In our data set we find that we have a relatively low rate of interhospital transfers for patients with a TS diagnosis of AIS mimics, highlighting the importance of accurate initial diagnosis. tPA administration in our group was higher than other reported hospitals in the literature.¹⁴,¹⁵ About half of patients receiving tPA required transfer to the hub hospital.

Our reported population has a relatively low proportion of patients with TIA and a higher proportion of tPA-treated AIS patients compared to population-based studies, both likely explained by our practice of encouraging our spoke site providers to consult our TS service primarily for patients potentially eligible for revascularization therapies. Thus, the results of this study are only be generalizable to other TS networks with similar consulting practices.

Our data supports prior studies which have shown that the most common AIS mimics are seizure, migraine, toxic metabolic disorders, psychiatric disorders, and peripheral CN palsies.⁸,¹² Our data also supports prior publications that found that predictors of AIS mimic during TS consultation include younger age and low NIHSS.⁸,¹⁶ Various clinically-based scores have been developed to attempt to differentiate AIS/TIA from stroke mimic, including the TeleStroke Mimic Score, FABS, and Khan Score. Our finding of lower age predicting stroke mimic replicates that same finding in all three clinical tools.⁸,^16–18 Nonetheless, it would be foolish to rely on younger age to discriminate between AIS and mimics given the rising prevalence of AIS in the young and the disproportionate impact of stroke on young adults in terms of disability and healthcare costs over a lifetime.¹⁹ Our finding of stroke mimic patients having lower NIHSS scores was replicated in both the TeleStroke Mimic score⁸,²⁰ and FABS.¹⁷

We have a higher rate of AIS mimics by TS diagnosis (44.3%) compared to previous studies, which report ranges of 16.7–33.6%⁷,¹⁶,²¹,²²; this could be due to a high tendency of our spokes to utilize the TS network for non-stroke cases. However, the prior studies only reported on the final diagnosis of patients in whom the TS consultant initially diagnosed AIS/TIA, whereas our study included consecutive patients, including those in whom the TS consultant initially diagnosed a stroke mimic. The proportion of patients in the current study with a final diagnosis of stroke mimic after initial TS diagnosis of AIS/TIA is 10.2%, and only 7.8% of patients who received tPA, comparable to earlier literature.

Limitations to our study include that it is a single-centre study involving one hub–spoke system and may not be representative of systems with different disease prevalence, stroke protocols, and consultant training. In addition, we had to exclude 14.9% of patients due to unclear documentation; in 73.2% of those cases, this was due to an unclear diagnosis by the discharging physician at spoke sites. Exclusion of unclear cases can skew our data. For example, Gargalas et al. notes that exclusion of cases with unclear documentation likely reduces the number of cases diagnosed with a functional disorder as providers can label diagnoses as unclear rather than as a diagnose a psychiatric disorder.²³,²⁴ The size of the study cohort precludes further analysis of the accuracy of stroke diagnosis for subcategories of AIS/TIA mimic. Final diagnoses are limited in that not all patients received a brain MRI; however, we believe that a clinical diagnosis in those in whom an MRI was not obtained is still relevant as it represents real-world practice. Due to a lack of systematic data collection of times of symptom onset, we cannot provide overall measures of symptom onset to time of consult. We also cannot provide overall estimates of AIS/TIA and mimic prevalence since the TS consultation process is not population-based but has a bias towards patients potentially eligible for revascularization.

Consultation for a high fraction of non-stroke cases could lower the overall accuracy if the stroke-trained consultant has an anchoring bias towards stroke diagnosis on initial evaluation. Conversely, accuracy may be artificially higher if the discharging physicians are biased towards agreeing with the specialist leading to higher concordance between TS consultation diagnosis and discharging diagnosis. Future studies could have a more in-depth review of diagnoses based on panel of experts that review the initial presentation and hospital course to determine a final diagnosis. Though more studies are needed to further assess the accuracy of TS in diagnosing specific AIS mimics, our research provides confidence in the utility of the TS network in accurately separating stroke from non-stroke.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Jason T Poon

Adam H de Havenon

Lee S Chung

References

Levine Steven

Gorman

“Telestroke”: the application of telemedicine for stroke. Stroke 1999; 30: 464–469.

Müller-Barna

Hubert

Boy

, et al. TeleStroke units serving as a model of care in rural areas. Stroke 2014; 45: 2739–2744.

Audebert

Kukla

Clarmann von Claranau

, et al. Telemedicine for safe and extended use of thrombolysis in stroke: the Telemedic Pilot Project for Integrative Stroke Care (TEMPiS) in Bavaria. Stroke 2005; 36: 287–291.

Richard

Wilcock

Schwamm

, et al. Assessment of telestroke capacity in US hospitals. JAMA Neurol 2020; 77: 1035–1037.

Katzan

Furlan

Lloyd

, et al. Use of tissue-type plasminogen activator for acute ischemic stroke: the Cleveland area experience. JAMA 2000; 283: 1151–1158.

Meyer

Raman

Hemmen

, et al. Efficacy of site-independent telemedicine in the STRokE DOC trial: a randomised, blinded, prospective study. Lancet Neurol 2008; 7: 787–795.

Mehta

Vora

Edgell

, et al. Stroke mimics under the drip-and-ship paradigm. J Stroke Cerebrovasc Dis 2014; 23: 844–849.

Ali

Viswanathan

Singhal

, et al. The TeleStroke Mimic (TM) – Score: a prediction rule for identifying stroke mimics evaluated in a telestroke network. JAHA 23 June 2014; 3: e000838. DOI: 10.1161/JAHA.114.000838.

Khanevski

Kvistad

Novotny

, et al. Incidence and etiologies of stroke mimics after incident stroke or transient ischemic attack. Stroke 2019; 50: 2937–2940.

10.

Wechsler

Demaerschalk

Schwamm

, et al. Telemedicine quality and outcomes in stroke: a scientific statement for healthcare professionals from the American Heart Association/American Stroke Association. Stroke 2017; 48: e3–e25.

11.

Sacco

Kasner

Broderick

, et al. An updated definition of stroke for the 21st century: a statement for healthcare professionals from the American Heart Association/American Stroke Association. Stroke 2013; 44: 2064–2089.

12.

Chernyshev

Martin-Schild

Albright

, et al. Safety of tPA in stroke mimics and neuroimaging-negative cerebral ischemia(Podcast). Neurology 2010; 74: 1340–1345.

13.

Goyal

Male

Al Wafai

, et al. Cost burden of stroke mimics and transient ischemic attack after intravenous tissue plasminogen activator treatment. J Stroke Cerebrovasc Dis 2015; 24: 828–833.

14.

Audebert

Kukla

Vatankhah

, et al. Comparison of tissue plasminogen activator administration management between telestroke network hospitals and academic stroke centers: the Telemedical Pilot Project for Integrative Stroke Care in Bavaria/Germany. Stroke 2006; 37: 1822–1827.

15.

Quain

Parsons

Loudfoot

, et al. Improving access to acute stroke therapies: a controlled trial of organised pre-hospital and emergency care. Med J Aust 2008; 189: 429–433.

16.

Ali

Hubert

Switzer

, et al. Validating the TeleStroke mimic score: a prediction rule for identifying stroke mimics evaluated over telestroke networks. Stroke 2018; 49: 688–692.

17.

Goyal

Tsivgoulis

Male

, et al. FABS: an intuitive tool for screening of stroke mimics in the emergency department. Stroke 2016; 47: 2216–2220.

18.

Khan

Chaku

Goehl

, et al. Novel algorithm to help identify stroke mimics. J Stroke Cerebrovasc Dis 2018; 27: 703–708.

19.

Madsen

Khoury

Leppert

, et al. Temporal trends in stroke incidence over time by sex and age in the GCNKSS. Stroke 2020; 51: 1070–1076.

20.

Geisler

Ali

Ebinger

, et al. Evaluation of a score for the prehospital distinction between cerebrovascular disease and stroke mimic patients. Int J Stroke 2019; 14: 400–408.

21.

Hand

Kwan

Lindley

, et al. Distinguishing between stroke and mimic at the bedside: the brain attack study. Stroke 2006; 37: 769–775.

22.

Hemmen

Meyer

McClean

, et al. Identification of nonischemic stroke mimics among 411 code strokes at the University of California, San Diego, Stroke Center. J Stroke Cerebrovasc Dis 2008; 17: 23–25.

23.

Gargalas

Weeks

Khan-Bourne

, et al. Incidence and outcome of functional stroke mimics admitted to a hyperacute stroke unit. J Neurol Neurosurg 2017; 88: 2–6.

24.

Reid

Currie

Baird

Non-stroke admissions to a hyperacute stroke unit. Scott Med J 2012; 57: 209–211.