Score for the Targeting of Atrial Fibrillation: A New Approach to Diagnosing Paroxysmal Atrial FibrillationSuissa L. · Mahagne M.H. · Lachaud S.
Centre Hospitalier de Nice – Hôpital Saint Roch, Unité Neurovasculaire, Nice, France Corresponding Author
Background and Purpose: Detecting paroxysmal atrial fibrillation (PAF) after ischaemic stroke is challenging. There are several methods to increase the detection rate of PAF, but it is first necessary to identify subgroups of patients at risk. In a previous study, we established a clinicoradiologic score that predicts atrial fibrillation (AF) in stroke patients. The purpose of the present study is to validate this score specifically for PAF patients. Methods: 500 consecutive ischaemic stroke patients were examined in our stroke unit. A blind evaluation of the STAF (score for the targeting of atrial fibrillation) was performed for each patient with or without AF. Firstly, we established the reproducibility of the STAF performance by comparing areas under the receiver operating characteristic curves in the preliminary and present studies. Secondly, to validate the predictive value of the STAF in occult AF, areas under the receiver operating characteristic curves were compared for each type of AF. Thirdly, the best threshold value was calculated. Results: AF was detected in 145 cases including 45% of paroxysmal forms. There is no significant score performance difference (p = 0.192) between the preliminary and prospective cohort areas under the receiver operating characteristic curves. This confirms the reproducibility of STAF performance. The area under the receiver operating characteristic curve for the PAF group was 0.907 versus 0.911 for the permanent AF group (p = 0.906). The diagnostic value of the STAF is as good in permanent as paroxysmal AF. In PAF, a STAF ≧5 has a sensitivity of 91% and a specificity of 77%. Conclusions: Due to its reproducibility and predictive value, the STAF can be used by neurologists as part of a novel diagnostic strategy for occult AF.
Copyright © 2011 S. Karger AG, Basel
The difficulty of diagnosing paroxysmal atrial fibrillation (PAF) leads to the underestimation of the prevalence of PAF and the underutilization of anticoagulants in the secondary prevention of ischaemic stroke . The cardio-embolic risk is the same whatever the type of atrial fibrillation (AF), permanent or paroxysmal . Permanent AF is easy to document, but diagnosis is challenging when the arrhythmia is paroxysmal. Routine diagnostic techniques such as 24-hour Holter monitoring have a low sensitivity. Increasing the frequency and duration of ECG monitoring improves the sensitivity of diagnosis [3,4,5,6,7], but such methods are infrequently employed due to their inaccessibility, high cost, and infeasibility with elderly patients who have had a severe stroke typical of a cardio-embolic origin . If the adage ‘seek and you shall find’ is correct, it does not apply to the daily reality that requires a specific justification for applying the aforementioned methods . So, it is necessary to select patients carefully before proposing such strategies. Our group developed the STAF (score for the targeting of atrial fibrillation) to identify, with an increased pretest probability, stroke patients who are suitable candidates for these techniques . The STAF is a clinicoradiologic score that allows the identification, with a sensitivity of 88% and a specificity of 89%, of AF patients among stroke patients. However, our previous study was retrospective and the STAF was established in AF patients and not specifically in PAF patients . Our present study therefore aims: (1) to confirm the reproducibility of the STAF in a prospective cohort, (2) to validate STAF performance in PAF, and (3) to find the best threshold value of the STAF for a sensitivity/specificity compromise in PAF.
We performed a prospective, monocentric study that included all consecutive patients admitted to the Nice stroke unit for documented acute cerebral infarction between November 2008 and May 2010. The patients received a clinical evaluation by a neurologist, including a TOAST classification . Patients with an incomplete clinical evaluation were excluded from the study. During the hospitalization, all patients had baseline ECG. Patients initially admitted to the intensive care unit underwent permanent ECG monitoring. All patients without documented AF using these methods had 24-hour Holter ECG. At 3 months, AF was documented by review of each patient’s medical history, initial and supplemental ECG, monitoring performed in the intensive care unit and 24-Holter ECG. Documented AF was classified as permanent or paroxysmal (terminated spontaneously and lasted less than 7 days).
The STAF was calculated in a blinded fashion at the end of the study using data collected during hospitalization. The required data included age, initial NIHSS score, left atrial diameter or surface area as measured by Doppler echocardiography according to European guidelines, and a vascular examination of the supra-aortic branches by Doppler ultrasound, computed tomography, magnetic resonance, or conventional angiography. The STAF was calculated using the criteria listed in table 1.
|Table 1. Criteria and scoring used in the STAF|
Validation of Score Reproducibility
The reproducibility of the discriminatory power of the STAF was assessed by comparison of the results of the present study with the previously published results . A receiver operating characteristic (ROC) curve of the STAF was established for the new prospective cohort. The area under the ROC curve and the diagnostic performance of the STAF (sensitivity, specificity) at the selected threshold value, as well as the corresponding 95% confidence intervals, were calculated and compared with previously published results. The threshold STAF value was set based on the ROC curve.
Validation of the STAF in PAF
The performance of the STAF in the paroxysmal model of AF was assessed by two approaches. In the first approach, the STAF ROC curves were plotted for paroxysmal and permanent AF. The area under each ROC curve and the corresponding 95% confidence interval were calculated and compared between groups. A STAF threshold value was also identified by this method for each type of AF. The sensitivity and selectivity, as functions of the STAF threshold value, were compared. In the second approach, the discriminatory power of each item in the STAF was examined by comparing its frequency (χ2 test) in each subtype of AF. A p value less than 0.05 was considered statistically significant.
Five hundred and eighteen consecutive patients were admitted to the Nice stroke unit with documented acute cerebral infarction over a period of 19 months. Eighteen patients (3%) were excluded from the study due to incomplete examination. The demographic characteristics and the aetiological distribution (based on the TOAST classification) of the present cohort and of the cohort from the preliminary study are presented in table 2 .
|Table 2. Baseline patient characteristics|
Three hundred and thirty patients (66%) were initially admitted to the intensive care unit and underwent ECG monitoring. The average duration of ECG monitoring was 129 h (4–864). At 3 months, AF was documented in 146 patients (29%). In 90%, AF was the only aetiology, and in 10% of cases it was associated with another potential cause of stroke (undetermined TOAST group). Arrhythmia was permanent in 55% of the patients and paroxysmal in 45%. The diagnostic methods for detecting PAF were ECG monitoring (57%), patient history (15%), supplemental ECG (14%), 24-hour Holter ECG (11%), and initial ECG (3%).
Figure 1 depicts a comparison of the independent STAF ROC curve for the prospective cohort with that from a previous publication . The areas under the curves showed no significant difference in the discriminatory power of the test between the two populations. A STAF threshold value of 5 or greater was set. For the chosen threshold value, the sensitivity was 92% (95% CI 86–96), and the specificity was 77% (95% CI 73–82).
|Fig. 1. Predictive value reproducibility of the STAF by comparison of areas under the receiver operating characteristic curves.|
As shown in figure 2, there was no difference between the discriminatory power for permanent and paroxysmal AF as a comparison of the areas under the curves yielded no significant difference. For the two types of AF, the STAF threshold value was set to 5 or greater. For the chosen threshold value, the sensitivity was 91% (95% CI 81–97) for detecting PAF, compared with a sensitivity of 92% (95% CI 84–97) for permanent AF. The specificity was 77% (95% CI 73–82) regardless of the type of arrhythmia. Sixty-nine percent (55 out of 80) of the false positives were patients older than 55 whose clinical evaluations were negative.
|Fig. 2. Independent comparison of STAF areas under the receiver operating characteristic curves in permanent and paroxysmal AF.|
Figure 3 shows the adequacy of the discriminatory capacities of each score item as a function of the type of AF.
|Fig. 3. Relevance of each item of the STAF in PAF.|
The difficulty in diagnosing PAF probably leads to the underestimation of PAF-associated ischemic stroke and the underutilization of anti-vitamin K therapy . We report that 29% of cerebral infarction cases were associated with AF, compared with 27% reported in our preliminary study . It is difficult to establish the precise prevalence of cardio-embolic stroke, which varies between 18 and 29% according to the cohort, but AF remains the leading cause of such stroke [13,14]. The proportion of paroxysmal arrhythmia reported in the literature is 40%, compared with 45% in our cohort [15,16]. The improvement of AF detection methods and the increased incidence of AF in the general population suggest an increase in the prevalence of AF-associated stroke at the expense of cryptogenic cerebral infarction .
Current routine detection methods have limited sensitivity. The 24-hour Holter monitoring identified 11% of new cases in our cohort, compared with 5% in the literature . It is not surprising to find a large proportion (57%) of new ECG monitoring diagnostics that, by increasing the recording duration, improve diagnostic efficiency [3,4,5,6,7].
The STAF, based on clinicoradiologic parameters, was conceived to assist the clinician in selecting patients for whom an aggressive search for AF should be considered . Our aim was to obtain a score easy to use. Every item composing the score is available in routine clinical practice. The reproducibility of the performance of the test is now established in a new prospective cohort. The discriminatory power of the score is the same regardless of the type of AF. We also confirmed the pertinence of each score item regardless of the type of arrhythmia. These results strengthen our strategy and refute the criticism concerning the applicability of the STAF to PAF based on a population with uncharacterized AF . The items that reflect initial severity and vascular evaluation are logically independent of the type of AF. The atrial echocardiographic criterion of the score is relevant whatever the type of AF. Indeed, atrial dilation is not only the consequence of atrial remodelling related to permanent arrhythmia. It can be correlated with causes and thrombo-embolic cardiologic factors of the fibrillation independent of its permanent or paroxysmal character [18,19]. While there is no significant difference in age between the two groups, this parameter seems to be less discriminatory for the paroxysmal form. It is not surprising when we consider that the paroxysmal form precedes the permanent form in the natural history of disease.
In clinical practice, the use of ECG recording techniques offers the practitioner tests with the requisite specificity (close to 100%) to justify the use of anti-vitamin K therapy . Their sensitivity is low but increases with the length of recording. Despite their costs, ECG monitoring techniques remain efficient . A STAF of 5 or greater, as reported here for the first time, adds sensitivity to the strategy. This sensitivity can be improved further by modifying the score threshold value, but at the cost of specificity. The specificity of the test does not permit the replacement of ECG recording methods. It is for this reason that with a STAF of 5 or greater, the majority of false positives are cryptogenic accidents in subjects over 55 years and for which the sensitivity of our AF diagnostic methodology was lacking. Thus, the STAF fits within a detection strategy proposed for a select population, allowing an increase in the efficiency of ECG monitoring tests. The selection of patients limited to cryptogenic cerebral infarction cases is one approach that often leads to the neglect of patients whose infarction is due to mixed causes (10% in our cohort). The detection of PAF by the STAF was validated for application to all cerebral infarction patients. Thus, the STAF contributes to optimization of the cost/benefit ratio, one of the factors limiting the routine use of these technologies. Other strategies have been proposed, such as the automated screening algorithm for PAF  or detection of premature atrial beats . These different methods might be combined in the future.
The STAF acts as the first link in a chain of paroxysmal AF detection efforts. It brings high sensitivity to ECG monitoring methods that have the specificity necessary for the prescription of anti-vitamin K therapy. The STAF rationalizes the non-routine use of ECG monitoring by increasing the pretest probability of a new diagnostic and thus improving the cost/benefit ratio. It can be used by neurologists in routine clinical practice. A validation of the diagnostic approach as well as an economic evaluation could help to propose a decision algorithm [9,20]. Such a future study will have to define the role of the STAF and the optimal duration of ECG monitoring.
Centre Hospitalier de Nice – Hôpital Saint Roch
Unité Neurovasculaire – Soins Intensifs
5, rue Pierre Devoluy, FR–6000 Nice (France)
Tel. +33 6 20 28 66 19, E-Mail email@example.com
Received: September 9, 2010
Accepted: December 16, 2010
Published online: February 23, 2011
Number of Print Pages : 6
Number of Figures : 3, Number of Tables : 2, Number of References : 22
Vol. 31, No. 5, Year 2011 (Cover Date: April 2011)
Journal Editor: Hennerici M.G. (Mannheim)
ISSN: 1015-9770 (Print), eISSN: 1421-9786 (Online)
For additional information: http://www.karger.com/CED