| Author: | Yang, Hualu |
| Title: | Development and validation of a prediction model for obstructive sleep apnea in stroke patients |
| Degree: | DHSc |
| Year: | 2024 |
| Subject: | Sleep Apnea, Obstructive Cerebrovascular disease -- Patients -- Rehabilitation Health risk assessment Hong Kong Polytechnic University -- Dissertations |
| Department: | Faculty of Health and Social Sciences |
| Pages: | 200 pages : color illustrations |
| Language: | English |
| Abstract: | Background: Obstructive sleep apnea (OSA) is one of common complications in stroke patients, leading to longer hospital stays, higher recurrence and mortality rates. It is critical for stroke patients with OSA to have timely diagnosis and receive effective treatments. Despite polysomnography (PSG) being the standard diagnostic tool for OSA, its substantial costs of manpower and equipment hinders its routine use in many hospitals. Some screening tools have been employed in clinical setting to detect individuals at high risk for OSA during the early stage, but few have demonstrated high efficacy with external validations, which limits their clinical applicability. Therefore, there is a need for a cost-effective, accurate, and user-friendly tool for early detection of OSA in stroke patients. Objective: The purpose of this study was to create and validate a model to predict the risk of OSA in stroke survivors. Method: A cross-sectional study was conducted on stroke patients admitted to a tertiary hospital's stroke unit from June 2022 to August 2023. Clinical data of 204 patients collected from June to December 2022 were utilized to develop and internally validate the models. Data from 136 patients collected from January to August 2023 were used for external validation. The primary outcome was the occurrence of OSA events within one week of stroke admission. The OSA was defined by a apnea hypopnea index (AHI) of 15 or higher per hour in PSG. There were 17 predictors selected in the models, including demographic factors (age, sex, smoking, alcohol consumption, history of diabetes), anthropometric measurements (body mass index, blood pressure, neck circumference), laboratory data (C-reactive protein), imaging data (infarction location), electrophysiology (heart failure), and disease presentations (snoring, respiratory arrest, Epworth sleep scale, dysphagia, and National Institutes of Health strokescale). The models were built using logistic regression, random forest, and decision tree methods. The models' performance was evaluated through calibration and discrimination measures, including the Hosmer-Lemeshow test, calibration plots, area under the curve of the receiver operating characteristic curve (AUC). Results: The prevalence of OSA among stroke patients was 59.8%. The logistic regression model showed that BMI, C-reactive protein, observed apnea, dysphagia and infarction on the brainstem were independent predictors for OSA in stroke patient (p<0.05). The model had satisfactory goodness-of-fit as shown in the calibration plot, and also good discrimination with an AUC of 0.819 (95% confidence interval [CI], 0.760-0.878). The optimum risk threshold of the logistic regression model was 0.658, and achieved the sensitivity of 0.696 and specificity of 0.802. The model had stable performance, with an internal validity of 0.75 in bootstraps for 1000 times. The AUCs of the models based on random forest and decision tree algorithms were 0.907 (95% CI: 0.863-0.7949), and 0.773 (95% CI: 0.708-0.835), separately. For external validation, 55.88% of the 136 stroke patients had OSA. The logistic regression model demonstrated strong discriminatory ability with an AUC of 0.780 (95% CI: 0.664-0.896). The random forest model and decision tree model had AUCs of 0.738 (95% CI: 0.614-0.860) and 0.672 (95% CI: 0.556-0.787), separately. The logistic regression model exhibited a slightly superior performance compared to the other two models with a higher AUC. Furthermore this model achieved stability in external validation, as indicated by the Delong test (D=0.594, p=0.553), same as decision tree model (D=1.492, p=0.138), whereas significant difference was found in the AUCs of the random forest model (D=2.546, p=0.012). Ultimately, the logistic regression model was chosen as the optimal OSA risk prediction model for stroke patients, and a web-based risk calculator (http://tt.fenghau.com/) was developed based on this regression model. This tool aims to improve accessibility and ease of use for healthcare professionals. Conclusion: This study developed three predictive models for OSA in stroke patients. The logistic regression model showed slight superiority over the other two models. The OSA prediction model for stroke patients holds promising practical implications for screening purposes. However, it is important to note that the external validation of the model is restricted to individuals within the same geographical location but at different time intervals, thereby limiting the generalizability of the validation outcomes. Future research endeavors should focus on conducting further verifications using larger sample sizes encompassing diverse regions to enhance the robustness and applicability of the model. |
| Rights: | All rights reserved |
| Access: | restricted access |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| 8005.pdf | For All Users (off-campus access for PolyU Staff & Students only) | 2.36 MB | Adobe PDF | View/Open |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13554