Strengths and limitations of this study

The study was set in a large stroke centre serving over 4000 inpatients, 100 000 outpatients and emergency visits annually.

This study built a visualised nomogram to identify moderate-to-severe sleep-disordered breathing in acute ischaemic stroke patients.

This study was conducted at a single centre.

A robust sample of 116 patients was used.

Introduction

Sleep-disordered breathing (SDB) is identified as a clinical syndrome, delineated by recurrent obstruction of the upper airway, which is commonly accompanied by snoring, sleep fragmentation and intermittent hypoxemia.1–4 The association between SDB and stroke has gained increasing attention.1 2 Obstructive sleep apnoea (OSA) is a primary manifestation of SDB in patients with acute ischaemic stroke (AIS).1 5 Studies have demonstrated that approximately 71% of acute stroke patients suffer from OSA.6 Moderate-to-severe SDB is a known risk factor for hypertension and atrial fibrillation, leading to cerebrovascular disease and vascular dementia.7 SDB can contribute to delay recovery and poor neurological outcomes, especially in patients with moderate-to-severe SDB.8 Furthermore, moderate-to-severe SDB is associated with early neurological deterioration, prolonged hospital stays and impaired cognition.6 9 Early detection of moderate-to-severe SDB in AIS patients requires early treatment and contributes to favourable outcomes.

However, moderate-to-severe SDB in patients with AIS has been under-detected. It is usually diagnosed using overnight polysomnography (PSG), which is less available and inconvenient to use. Additionally, lengthy outpatient waiting lists for PSG cause a delay in diagnosis. Previous screening scores, such as STOP-Bang questionnaire (SBQ), was underpowered to detect many moderate-to-severe SDB patients who fail to manifest typical symptoms like disruptive snoring and daytime sleepiness. Currently, no moderate-to-severe SDB screening model has been developed and validated for AIS population. Therefore, it is imperative to develop a model that can identify moderate-to-severe SDB conveniently and effectively in AIS patients and aid clinical screening.

Materials and methodsPopulation

We consecutively enrolled 116 participants with AIS at the Department of Neurology in Shanghai Tenth People’s Hospital between May 2020 and August 2021. 112 AIS patients were confirmed using MRI, and 4 were diagnosed using CT 48 hours after onset, as 1 patient could not tolerate cranial MRI examination due to claustrophobic symptoms, and 3 had stents or plates installed. The exclusion criteria were as follows: (1) individuals aged 18 years or younger, (2) decreased consciousness level, (3) a history of chronic respiratory disease or any acute cardiac or respiratory condition that leads to hypoxia, (4) severe behavioural disorders and aphasia and (5) underwent emergency surgical procedures. All enrolled patients completed PSG within 7 days of hospitalisation, and they were then divided into the no-to-mild SDB group (apnoea-hypopnoea index (AHI) <15) or moderate-to-severe SDB group (AHI≥15). The patient selection flowchart is presented in figure 1.

Figure 1

Flow chart demonstrating details of various steps of the randomised controlled study. AHI, apnoea-hypopnoea index; PSG, polysomnography; SDB, sleep-disordered breathing.

Clinical, demographic and imaging characteristic equations

Data on age, sex, body mass index (BMI) and neck circumference were collected or measured at admission. Obesity was defined as a BMI of ≥28 kg/m2. Patient medical histories were documented, including the presence of hypertension, diabetes mellitus, ischaemic heart disease, atrial fibrillation, and previous ischaemic stroke or transient ischaemic attack (TIA). The ischaemic sites were determined by CT or MRI within 7 days after ischaemic stroke onset and categorised as lobar (frontal, parietal, temporal, insular or occipital), subcortical (thalamus, basal ganglia, internal capsule, corona radiata or corpus callosum), brainstem (midbrain, pons or medulla) and cerebellum. We also assess the pathogenetic types of stroke with Trial of Org 10 172 in Acute Stroke Treatment (TOAST classification) (large-artery atherosclerosis, cardioembolism, small-vessel occlusion, stroke of other determined aetiology, stroke of undetermined aetiology).

Blood test

The blood test parameters included white blood cell and red blood cell counts, as well as hemoglobin, fibrinogen, C reactive protein, platelet, triglyceride, total cholesterol, high-density lipoprotein, low-density lipoprotein, uric acid, fasting blood glucose, glycated haemoglobin and homocysteine levels. Additionally, the C reactive protein to albumin ratio; fibrinogen to albumin ratio, and the levels of tumour markers, including carcinoembryonic antigen (CEA), carbohydrate antigen 199(CA199), carbohydrate antigen 724 (CA724), neuron-specific enolase (NSE) and alpha-fetoprotein(AFP), were recorded.

Overnight PSG

All patients underwent PSG overnight for at least 7 hours within 7 days of hospitalisation, which involved the use of an electroencephalogram, a genioglossus electromyogram, a thermistor to monitor oral and nasal flow, an ECG, a thoracoabdominal respiratory movements recording and an oxygen saturation measurement. Data were automatically generated by the system and scored using a professional sleep monitor. Our PSG results are initially subjected to automated scoring by the machine. Subsequently, a meticulous manual review is performed to ensure precision and reliability. This manual review process is conducted under the supervision of a specialist in sleep breathing disorders, strictly adhering to the guidelines outlined in the American Academy of Sleep Medicine Scoring Manual. This dual-step approach, which combines automated scoring with specialist oversight, is integral to upholding the utmost standard of accuracy in our analyses. We used standardised criteria to score sleep parameters and respiratory events. Apnoea was identified when there was a cessation of airflow lasting at least 10 s, whereas hypopnoea was determined by a reduction in airflow of 30% or more for a minimum duration of 10 s, coupled with either a reduction in oxygen saturation greater than 3% or an arousal. The AHI was calculated as the apnoea score plus the hypopnoea score per hour of sleep. Apnoea severity was measured by using the AHI. Patients with an AHI of <15 were categorised as having no-to-mild SDB, and those with an AHI of ≥15 were classified as having moderate-to-severe SDB. Sleeping scales including SBQ were assessed by a professional doctor.

Statistics

Descriptive statistics were used to summarise baseline characteristics of patients. We used the Kolmogorov-Smirnov to detect normal distribution among the quantitative variables. Continuous variables following a normal distribution were described using means±SD, while those not normally distributed were reported as medians. The t-test was used for continuous data with normal distribution, while the Mann-Whitney U test was applied to compare variables with a non-normal distribution, and the χ2 test was used for categorical measures. Multivariate logistic regression was used to determine the predictors of risk factors. All variables with p<0.05 in the univariate analysis were included in multivariate analyses. A nomogram was generated using multivariate logistic regression analysis to predict the factors influencing moderate-to-severe SDB in patients with ischaemic stroke using a backward stepwise method. The strength of the association was assessed using ORs and the corresponding 95% CIs. The probabilities of entry and removal were set to 0.05 and 0.10, respectively. Variables with p<0.1 were entered for analysis in the R language to establish the nomogram of the screening model. The predictive accuracy of the nomogram model was assessed by calculating the area under the receiver operating characteristic curve (AUROC). For internal validation, the bootstrap technique, using 1000 bootstrap samples, was employed. The corrected concordance index (C-index) was calculated, with a range of 0.5–1.0 and a higher C-index representing a more accurate power of the model. We compared our model with the traditional SBQ to examine the accuracy of our prediction model. Finally, the calibration curves were plotted based on the overall nomogram score. Data were analysed using SPSS software (V.22 IBM Corporation, New York, USA) and R software (V.4.1.1 2019, The R Foundation for Statistical Computing Platform).

Patient and public involvement

Patients and the public were not involved in the design, conduct or dissemination of this study.

ResultsDemographic data, clinical characteristics and univariate analysis

Of the 116 patients enrolled in this study, 53 (45.7%) had moderate-to-severe SDB with an AHI of ≥15. The cohort was predominantly male (86.2%), with an average age of 62 (55.3–66.8) years, and the median National Institutes of Health Stroke Scale (NIHSS) score of 2 (1–4). The highest NIHSS score was 12. The mean BMI was 26.2±3.4 kg/m2, and the median neck circumference was 40 (39–42) cm. 14 patients (12.1%) presented with wake-up stroke (WUS), and 15 received thrombolytic therapy (12.9%). Of the enrolled patients, 101 patients had hypertension (87.1%) and 60 had diabetes (51.7%). In the univariate analysis, BMI, neck circumference, NIHSS score, presence of brainstem infarctions, CEA, CA724, and NSE levels were significantly different between the moderate-to-severe SDB and no-to-mild SDB groups (p<0.05) (table 1). The TOAST classification was negative (p=0.287), so it was not included in the model.

Table 1

Characteristics and univariate comparison of no-to-mild SDB group (AHI<15) and moderate-to-severe SDB group (AHI≥15)

Multivariate analysis of the development cohort

The binary logistic regression model included variables that were significantly different in the univariate analysis. BMI, CEA and CA724 levels were not significant variables in the binary logistic regression model and were therefore eliminated. The final model included five potential predictors: NIHSS score (OR=1.20; 95% CI 0.98 to 1.47), neck circumference (OR=1.50; 95% CI 1.16 to 1.95), presence of WUS (OR=21.91; 95% CI 3.08 to 156.05), NSE level (OR=1.27; 95% CI 1.05 to 1.53) and presence of brainstem infarctions (OR=4.21; 95% CI 1.23 to 14.40) (table 2). The AUROC of our model was 0.872, with a sensitivity of 75.5% and a specificity of 85.7%, outperforming the predictive accuracy of SBQ (ROC 0.733, sensitivity 88.7%, specificity 46.0%) (figure 2).

Table 2

Multivariate analysis for predictors of moderate-to-severe sleep-disordered breathing in the patient with acute ischaemic stroke

Figure 2

Receiver operating characteristic (ROC) analysis of the predictive model (combination of National Institutes of Health Stroke Scale baseline neck circumference, wake-up stroke, neuron-specific enolase and brainstem stroke) compared with STOP-Bang questionnaire (SBQ).

Performance of the nomogram to evaluate the likelihood of moderate-to-severe SDB in patients with AIS

A nomogram was developed to evaluate the likelihood of moderate-to-severe SDB by incorporating five independent variables. The integral values for each factor used to calculate the total score are shown in figure 3. The calibration curve of the nomogram demonstrated good consistency in this cohort (online supplemental figure 1). The results of the 1000 bootstrap samples had a C-index of 0.849, suggesting the model’s good discrimination ability.

Figure 3

Nomogram for predicting the probability of moderate-to-severe sleep-disordered breathing (SDB) in the patient with acute ischaemic stroke. To use the nomogram, it is first needed to determine the position of each variable on its axis and then plot a line with the specified value on the axis to determine the score of each variable. Next, the total score of all predictor variables is calculated below the nomogram to determine the probability of moderate-to-severe SDB. NIHSS, National Institutes of Health Stroke Scale; NSE, neuron-specific enolase; WUS, wake-up stroke.

Discussion

In this retrospective cohort study, we developed and validated a nomogram that incorporates five clinical variables to assess the likelihood of moderate-to-severe SDB in mild AIS patients. Nomograms are widely used in research for risk prediction, calculating individual risk by considering relevant population-specific factors.10–12 Our results showed that the screening tool has good discrimination ability and a satisfactory goodness‐of‐fit. Compared with the SBQ, our model showed superior accuracy and practicability in assessing the likelihood of moderate-to-severe SDB in AIS.

SDB is a treatable condition, and once diagnosed, patients may benefit from continuous positive airway pressure treatment. However, it is challenging for clinicians to detect moderate-to-severe SDB in patients with AIS, as not all patients can undergo overnight PSG during hospitalisation. The SBQ is widely used in otolaryngology clinics, and it appears to be a convenient tool for evaluating SDB.13 However, this scale is not derived for the stroke population and does not consider stroke-associated variables. In addition, stroke patients with SDB often remain asymptomatic. Therefore, it is crucial to develop a more reliable tool specifically for stroke population.

Our study found that 45.7% of the participants had moderate-to-severe SDB, defined as an AHI of ≥15 per hour. Five variables were independently associated with moderate-to-severe SDB, including presence of WUS, brainstem infarction, neck circumference, NIHSS score and NSE level. A prior study with 95 TIA and mild to moderate stroke patients showed that WUS was associated with moderate-to-severe OSA and worse short-term outcomes.14 We also found a strong correlation between WUS and SDB, with patients with WUS having an approximately 21.91 times higher risk of SDB when compared with those without WUS. A possible explanation for this might be the occurrence of severe hypoxia at night, which contributed to the WUS in patients with SDB. Therefore, it is critical to screen for SDB, particularly moderate-to-severe SDB, in patients with WUS to allow for earlier treatment.

In addition, we observed a higher prevalence of brainstem infarction in the moderate-to-severe SDB group than in the no-to-mild SBD group in our study. However, the correlation between SDB and ischaemic stroke site remains controversial, with some studies with small sample sizes failing to demonstrate statistically significant associations between infarction sites and SDB. Conversely, a result reported a strong correlation between brainstem infarctions and SDB.15 Another study reported an association between acute brainstem infarction and the presence and severity of SDB,16 consistent with our results. Respiratory rhythm is generated by a network of neurons located in the ventrolateral medulla. The underlying rhythmic pattern of the rhythmic drive is determined by the interaction between neurons and several respiratory nuclei situated in the medulla and pons. Brainstem infarction involving these nuclei results in a non-rhythmic or allorhythmic respiratory pattern, leading to SDB. Additionally, brainstem infarctions commonly weaken pharyngeal dilator muscles, leading to upper airway resistance and an elevated risk of SDB. Obesity is known to play a crucial role in the pathogenesis of SDB, with BMI can be used as an indicator of obesity. Therefore, it is reasonable to assume that BMI is an independent risk factor for SDB. In our study, univariate analysis showed that BMI was associated with SDB, as well as neck circumference. However, the multivariate analysis indicated that neck circumference was independently associated with SDB, rather than BMI. This suggests that neck circumference might serve as a mediator between BMI and SDB. Therefore, we excluded BMI from the model and included neck circumference instead. Consistent with our findings, a Finnish study reported positive associations between NIHSS scores at admission and discharge with SDB, mirroring the outcomes observed in our cohort.17 In our study, minor strokes dominated the cohort, with a median NIHSS score of 2.1–4 Good compliance with overnight PSG completion in minor stroke might explain this finding. Our findings also indicated that even in the minor stroke population, the correlation between the NIHSS score and SDB remained. We further found that NSE plays an important role in assessing moderate-to-severe SDB in patients with stroke. Elevated NSE level is considered a peripheral indicator of brain damage in OSA.18 The nocturnal hypoxia may contribute to increased serum NSE level.

Our study has limitations. First, despite having a larger sample size than many comparable studies, the completion rate (116 out of 326 enrolled) fell short of the number initially targeted to achieve optimal statistical power (150 participants). This limitation is crucial for the interpretation of our findings and highlights the importance of conducting future research with more extensive cohorts to corroborate our results. Second, in our study, the inclusion of patients predominantly with mild to moderate ischaemic strokes (maximum NIHSS score of 12) reflects a selection bias, primarily due to safety concerns associated with conducting PSG in individuals with more severe conditions. However, it indicates that our findings are most applicable to patients with an NIHSS score of 12 or below, a subgroup that may derive greater benefits from early identification and intervention. In addition, despite having a well-defined internal validation process, we did not have an external validation process. Therefore, a multicentre, large-scale validation study is required. In our ongoing work, we plan to develop a user-friendly interface to transform the nomogram into a small programme, making it more accessible and efficient for both healthcare professionals and patients. In the interim, we will indeed convert the nomogram into a questionnaire format to facilitate ease of use (online supplemental figure 2).

Conclusion

Our study presents a practical nomogram model to assess the probability of moderate-to-severe SDB in patients with AIS. This nomogram is specifically tailored for a stroke population, offering a valuable tool for early detection of moderate-to-severe SDB and potentially leading to improved outcomes.

Data availability statement

Data are available upon reasonable request.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

This study involves human participants and was approved by Ethics Committee of Shanghai Tenth People’s Hospital (SHSY-IEC-4.1/21-304/01). Participants gave informed consent to participate in the study before taking part.

Acknowledgments

We are grateful to Dr CX Shen and Dr X Li (from Department of Pneumology, Shanghai Tenth People’s Hospital, Tongji University) for their PSG diagnosis.