Pre‐screening of sleep‐disordered breathing after stroke: A systematic review

Abstract Objectives Sleep‐Disordered Breathing (SDB) is frequent in stroke patients. Polysomnography (PSG) and cardiorespiratory polygraphy are used to confirm SDB, but the need for PSG exceeds the available resources for systematic testing. Therefore, a simple and robust pre‐screening instrument is necessary to identify the patients with an urgent need for a targeted PSG. The aim of this systematic review was to identify and evaluate the available methods to pre‐screen stroke patients possibly suffering from SDB. Materials and Methods Eleven studies out of 3,561 studies met the inclusion criteria. The selected studies assessed the efficiency of seven instruments based on the data acquired clinically or by inquiries (Berlin Questionnaire, Epworth Sleepiness Scale, SOS, Modified Sleep Apnea Scale of the Sleep Disorders Questionnaire, STOP‐BANG, Four‐variable Screening Tool and Multivariate Apnea Index) and three physiological measures (capnography, nocturia, nocturnal oximetry). The instruments were used to predict SDB in patients after acute or subacute stroke. Either PSG or cardiorespiratory polygraphy was used as a standard to measure SDB. Results No independent studies using the same questionnaires, methods or criteria were published reducing generalizability. Overall, the questionnaires were quite sensitive in finding SDB but not highly specific in identifying the non‐affected. The physiological measures (capnography) indicated promising results in predicting SDB, but capnography is not an ideal pre‐screening instrument as it requires a specialist to interpret the results. Conclusions The results of pre‐screening of SDB in acute and subacute stroke patients are promising but inconsistent. The current pre‐screening methods cannot readily be referred to clinicians in neurologic departments. Thus, it is necessary to conduct more research on developing novel pre‐screening methods for detecting SDB after stroke.

An increasing number of researches address the need for systematic SDB screenings after stroke. Untreated sleep disorders can increase the risk for recurrent strokes (Yaggi et al., 2005), whereas treatment of SDB with continuous positive airway pressure may reduce mortality after stroke (Martínez-García et al., 2009). Moreover, adherence to sleep apnea treatment reduces the mortality rate as compared to untreated patients (Ou, Chen, Zhuo, Tian & He, 2015).
However, some disagreement remains (McEvoy, Antic, Heeley, Luo & Ou, 2016). Sleep apnea is listed as a risk factor as well as a consequence of stroke in the European guidelines for cerebrovascular disease (ESO, 2008). Therefore, recognition and treatment of SDB after stroke constitute an important part of the secondary prevention and rehabilitation process. Early identification and treatment of SDB could enhance rehabilitation and decrease the patients' time in hospital as well as increase the quality of life.
A polysomnography (PSG) or cardiorespiratory polygraphy are standard methods needed to diagnostically assess the severity of SDB. They are used to measure the Apnea-Hypopnea Index (AHI) indicating the mean number of apnea or hypopnea events per hour. Unfortunately, there are more stroke patients than resources available for systematic PSG testing. SDB prescreening after stroke can also be considered an action in the prevention of recurrent stroke which could be beneficial in reducing disability and mortality in the long run. Thus, a simple targeted SDB pre-screening method, which can potentially identify patients who should undergo more formal PSG, is needed. Early identification and treatment can boost rehabilitation, reduce time spent in hospitals and prevent recurrent strokes. As a result, ideally the economic burden that public health care poses on society could be reduced and great financial savings made.
F I G U R E 1 Flow chart of the articles reviewed, excluded and analyzed for the present systematic review Records excluded (n = 2327) Reasons: ICU setting, chronic stroke, reviews, animal studies, pediatric, diagnosed sleep apnea, case studies, pain, depression, drugs, patients in unstable medical condition, heart stroke Full-text articles assessed for eligibility (n = 37) Full-text articles excluded (n = 26) Studies included in qualitative synthesis (n = 11) Studies included in quantitative synthesis (meta-analysis) (n = 0) Studies that did not assess the diagnostic accuracy of the index test against reference test (n = 23) Studies that did not use valid reference test to compare the index test with (n = 2) Studies with full text not found in English (n = 1) Identifying predictive signs of SDB could help in finding the patients who benefit from the administration of PSG. Ideal SDB pre-screening should be simple and fast for the medical or nursing personnel to administer, and it should not require a specialist's interpretation. The method should be sensitive in finding the patients at risk. Specificity could then be confirmed with more thorough PSG testing. The present systematic review aims to assess and evaluate current literature on existing SDB pre-screening methods after acute or subacute cerebrovascular stroke and the predictive power of such methods.

| ME THODS
A systematic review was conducted according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guideline (PRISMA guideline). This systematic approach was selected because the focus was solely on experimental articles, and the aim was to include all available experimental evidence. The The search terms consisted of the following index terms "sleep," "screening" and "stroke," and additional search terms related to these index words such as "assessment," "evaluation," "questionnaire," "monitor," "measure," "quality," "scale," "polygraph," "polysomnography," "actigraph," "actometer," "stroke," "cerebral infarct," "cerebral hemorrhage or cerebral haemorrhage," "TIA or transient ischemic attack" or "cerebral ischemia" (Figure 1). No additional filters were included. In addition, the reference lists of the selected articles were checked.
The inclusion criteria for the studies consisted of the following: • The study was conducted on acute (within an initial stay at the hospital due to the first onset of stroke) or subacute (within 1 year after stroke) cerebrovascular stroke patients (transient ischemic attack, cerebral infarct or intracerebral hemorrhage); • The study used a pre-screening method to predict SDB with calculated sensitivity and specificity; • The study used either PSG or cardiorespiratory polygraphy as a standard to measure AHI and to compare the index test with; and • The full text of the study was written in English.
First, the abstracts of the articles were reviewed by two researchers (MT, AH) blindly and independent of each other. The other members of the research group (JP, ER) were consulted if any disagreements occurred. The final decision required all members' full agreement.
Second, the data were extracted by two reviewers (MT, AH) in collaboration. The final extraction included the entire research group (MT, AH, JP, ER). Study characteristics, sensitivities and specificities and negative and positive predictive values (NPV, PPV) were collected from each paper as thoroughly as they were reported. Only reported results from each paper were included excluding any data requiring extrapolations or derivations from graphs or tables. The results from the studies with acute and subacute strokes were pooled together in our analysis. Studies with chronic phases were excluded.
The internal and external validities were assessed for each article according to Cochrane Methods Group on Screening and Diagnostic Tests guideline (Reitsma et al., 2009). Internal validity consisted of the following factors: valid standard test, definition of AHI by a standard test (full polysomnography or cardiorespiratory polygraphy), blind execution of tests, verification bias, and independent analysis of standard and index tests. External validity consisted of the following factors: disease spectrum, background information, cutoff values, missing data, index test completion, and the method for subject selection.
More specifically, seven studies (

| Questionnaires or a prediction model
The Epworth Sleepiness Scale (Johns, 1991) assesses daytime sleepiness by evaluating the tendency to fall asleep in given situations. ESS was used in two studies (Camilo et al., 2014;Srijithesh et al., 2011).
SOS score (Camilo et al., 2014) combines the elements of BQ (Netzer et al., 1991) and ESS (Johns, 1991) by adding both together with a modified scoring. It was demonstrated in two studies (Boulos et al., 2015;Camilo et al., 2014).
Sleep Disorders Questionnaire (SDQ-SA) (Douglass et al., 1994) is a basic sleep apnea questionnaire used commonly in SDB prescreening in clinical research. One study (Bassetti et al., 2006) used SDQ-SA (Douglass et al., 1994) in combination with ESS (Johns, 1991) to predict SDB so that Probable Sleep Apnea (P-SA) was defined by ESS > 10 or SDQ-SA ≥ 32 in women and ≥36 in men. screening tool was assessed in one study (Boulos et al., 2015).
Multivariate Apnea Index index (Maislin et al., 1995) has been used in predicting obstructive sleep apnea. It uses the patient's age, sex, and BMI in the prediction model. MAP was used in sleep apnea evaluation in one study (Broadley et al., 2007).

| Physiological measures
Three studies (Aaronson et al., 2012;Chen et al., 2011;Dziewas et al., 2005) used physiological methods for SDB pre-screening and compared the results with a standard sleep apnea test. One study (Aaronson et al., 2012)

| Questionnaires or prediction model
The definitions of cutoff values for AHI between the studies were highly non-uniform. For example, some studies set the cutoff for AHI to 10 and some to 15. Therefore, we divided the studies into four groups according to the AHI cutoffs the authors had used as follows: AHI ≥ 5/hr, AHI ≥ 10-15/hr, AHI ≥ 20, AHI ≥ 30 (Table 3).
Three studies (Elkholy et al., 2012;Katzan et al., 2016;Srijithesh et al., 2011) reported the results with a cutoff of AHI ≥ 5/hr. One study (Srijithesh et al., 2011) tested BQ, BQ and ESS separately and in combination but none of them showed particularly high specificities or sensitivities in predicting SDB. One study (Elkholy et al., 2012) also assessed BQ and found it to be highly specific but not
All studies that used questionnaires or prediction models reported the results with AHI cutoff of either 10 or 15 (AHI ≥ 10-15/ hr) (Bassetti et al., 2006;Boulos et al., 2015;Broadley et al., 2007;Camilo et al., 2014;Elkholy et al., 2012;Katzan et al., 2016;Kotzian et al., 2012;Srijithesh et al., 2011). The sensitivities ranged from 52.0% to 100% and specificities from 14.0% to 100%. The most sensitive instrument in predicting SDB was the 4V questionnaire with 97.0% sensitivity. The specificity, however, was only 24% and the area under curve (AUC) was poor, that is, 67.7%. Hence, this instrument recognizes the affected ones well but those non-affected poorly. The most specific instrument in the moderate SDB group was BQ with 86% specificity. Sensitivity, however, was only 56% indicating that nearly half of the affected ones remain unidentified. One study used AHI cutoff point ≥20/hr (Katzan et al., 2016) (Table 3).
The same trend continued in the category with the highest cutoff for AHI (AHI ≥ 30/hr) as SOS-score, BQ, ESS, and STOP-BANG, and its derivatives indicated good sensitivities to predict SDB but low specificities (Table 3) (Camilo et al., 2014;Elkholy et al., 2012;Katzan et al., 2016).

| Physiological measures
Three studies (Aaronson et al., 2012;Chen et al., 2011;Dziewas et al., 2005) used physiological measures (capnography, nocturnal oximetry, nocturia) to test their power to predict SDB (Table 3). The most sensitive and specific of these was the capnography (Dziewas et al., 2005) measurement with sensitivity, specificity, and negative and positive predictive values of 87.0%, 100%, 86.0%, and 100%, respectively. The corresponding values of the nocturnal oximetry (Aaronson et al., 2012) were 77.0%, 100%, 83.0%, and 100%, and nocturia (Chen et al., 2011) 68.0% and 59.0% with no reported NPV and PPV. There were no reports on the demands for resources such as personnel or time of physiological measures compared to standard tests for SDB (polysomnography or cardiorespiratory polygraphy).

| D ISCUSS I ON
To our knowledge, this is the first systematic review on SDB prescreening methods in acute and subacute stroke patients. Prescreening methods for detecting SDB after stroke have not been studied extensively, as only eleven different pre-screening methods for Sleep-Disordered Breathing after acute and subacute stroke were identified. The results show that some pre-screening methods might have the potential to identify patients suffering from SDB before polygraphy for targeted testing. Questionnaires are more desirable pre-screening methods due to their simplicity as they can be self-answered or filled in by a nurse on the basis of an interview at a stroke unit. Even if the questionnaires are easy and fast to administer, their predictive value was proved to be poor, and they cannot be clinically recommended for SDB screening after stroke. The physiological measures (capnography, nocturnal oximetry) produced the best predictive results but their usability for screening is greatly diminished due to their resource needs, that is, equipment, time-consuming overnight monitoring, and specialist interpretation on data.
Extensive research has emphasized that SDB is highly prevalent after stroke and there might even be a causal relationship between SDB and stroke (Bassetti et al., 1996;Cam et al., 2013;Gao et al., 2010;Hermann & Bassetti, 2009;Martínez-García et al., 2009;Ou et al., 2015;Sahlin et al., 2008;Yaggi et al., 2005). In fact, sleep apnea is generally recognized as an independent risk factor for stroke (Yaggi et al., 2005). The problem is in identifying those at risk because the resources to screen for SDB are limited in neurology de- For pre-screening SDB in acute and subacute stroke patients, questionnaires can be conducted quickly and the results can be assessed immediately. However, the existing literature does not fully succeed in reassuring the functionality of questionnaires as a SDB pre-screening method, because the diagnostic accuracies were altogether modest. For example, although performing well in identifying the SDB positive patients, only every fourth patient would be accurately diagnosed as non-affected with a 4V questionnaire due to the low specificity of the test resulting in a high number of false positives. This is not useful in decreasing the work of specialized physicians and therefore not a very practical pre-screening instrument.
Moreover, there was some discrepancy in the results of the Berlin Questionnaire (BQ) (Netzer et al., 1991), as they varied tremendously between studies which affect detrimentally the creditability of its use in SDB screening in neurologic patients. For example, the majority of studies using BQ (Boulos et al., 2015;Camilo et al., 2014;Kotzian et al., 2012;Srijithesh et al., 2011) concluded that BQ is not sufficient in predicting SDB at the moderate severity level, while one study (Elkholy et al., 2012) recommended the opposite.
Also, when looking at the specificities between these studies, it is impossible not to question the plausibility of the results as they vary considerably ( Table 3). The heterogeneity of the methods and different cutoff points for moderate AHI prevented us from performing a meta-analysis on the pre-screening methods for SDB after stroke.
To conclude, no plausible and pragmatic tool for clinical pretesting of SDB exists according to our systematic review. Currently, no specific SDB pre-screening method can be referred to clinicians working in neurologic departments. Thus, polysomnography or cardiorespiratory polygraphy is still clinically needed when suspecting SDB in stroke patients. Still, the high SDB prevalence among stroke patients remains and the physicians in stroke units need to discretionarily prescribe further sleep testing to susceptible patients. More research is needed to find an optimal pre-screening instrument for clinical practice to identify SDB patients after stroke.

ACK N OWLED G M ENT
We thank information specialist Miia Ulanen, BBA, Medical Library Hospital District, Finland).

CO N FLI C T O F I NTE R E S T
Mrs. M. Takala reports no disclosures relevant to the manuscript.