Figure 1.
Figure 2.
Figure 3.
Figure 4.
Database search and selection criteria
| Electronic database |
|
| Inclusion criteria |
|
| Exclusion criteria |
|
Search strategy
| Population | Studies using physiological data to build sleep apnea classi^ication algorithms |
| Comparison | Different models and their utility for clinical intervention |
| Outcome | Ability to detect or predict sleep apnea, sleep arousals, respiratory events during sleep |
| Study type | Quantitative study |
| Keywords | Sleep apnea, recommender system, polysomnography, machine learning, artificial intelligence |
Treatment options for SA
| Author | Treatment | Observations | Improvements |
|---|---|---|---|
| Boisteano, et al. 2009 [33] | fixed CPAP, ACPAP | ACPAP effectively lowers pressure. It is a long-term treatment option. | Intelligent CPAP with remote monitoring can reduce frequent doctor visits, reducing overall cost. |
| Penzel et al. 2011 [34] | CPAP | OSA affects cardiovascular, respiratory regulation during sleep | CPAP affects cardiovascular coupling during deep sleep; baroreflex sensitivity response varies across sleep stages. |
| Jané, 2014 [30] | CPAP, adaptive pressure algorithm, M-Health, Tele-Health | OSA disrupts airflow during sleep; SpO2is also reduced. | Standard treatment option is CPAP, advanced CPAP with adaptive pressure settings. |
| Aurora et al. 2016 [36] | ASV for CSA patients | Cardiac mortality rate is high for LVEF >= 45%, moderate-severe CSA. | ASV suggested for patients with LVEF >45% and mild-moderate CSA. |
| Senavongse et al. 2017 [31] | A low-cost functional prototype is designed. | Modern CPAP machine is not affordable to everyone nowadays. | The functional prototype demonstrates accurate measurements and improves OSA, snoring treatment. |
| Webster et al. 2018 [37] | Novel SA treatment device consisting of mask, hose pipe, CO2 chamber. | Device automatically adjusts rebreathed air to reduce apneas. | No need for CPAP |
| Amrulloh et al. 2019 [32] | OCPAP controller developed using LabView software | OCPAP adapts air pressure based on respiratory needs | Low-cost treatment option that reduces dependency on CPAP. |
| Daga et al. 2021 [35] | MAD and yoga exercise | Sleep quality is assessed before and after MAD surgery. | MAD group showed immediate improvements while the yoga group showed sustained improvements over a long time period. |
DL algorithms for SA diagnosis
| Author | Bio signal | DL Algorithm | Performance | Type of Classification | ||
|---|---|---|---|---|---|---|
| Accuracy (%) | Sensitivity | Specificity | ||||
| Huttunen, 2023 [25] | SpO2, PR, ECG | CNN, RG | – | – | – | detection |
| Sharma et al., 2022 [49] | SpO2, PR | CNN | 93.4% | – | – | detection |
| Strumpf et al., 2023 [56] | SpO2, HR | ANN | 91% | 0.83 | 0.76 | multiclass classification |
| Hemrajani et al., 2023 [57] | ECG | RNN, LSTM, GRU. | 89.5% RNN; 90% LSTM; 90.5% GRU | – | – | classification |
| Korkalainen et al., 2021 [26] | EEG, SpO2 | CNN, RNN | hazard ratio = 1.14 (p = 0.39) for mild OSA | hazard ratio = 1.59 (p < 0.01) for moderate OSA | hazard ratio = 4.13 (p < 0.01) for severe OSA | estimation |
| Liu et al., 2020 [45] | EEG, SpO2 | CNN, RF | AUROC = 0.95 | AUPRC = 0.552 | – | detection |
| Mitri et al., 2017 [46] | Nasal pressure, CPAP pressure readings | HTM, NAB | – | – | – | anomaly detection |
| Sun et al., 2019 [55] | EEG | RNN | – | – | – | binary classification |
| Yung et al., 2020 [58] | ECG | 1D-CNN | 89% | – | – | detection |
| Zarei et al., 2021 [29] | ECG | CNN-LSTM | 97.21% | 94.41 % | 98.94% | detection |
Various Data sets available for SA
| Data Set | Sample Size | Types of Signals | Time Range | Male | Female |
|---|---|---|---|---|---|
| UCDDB data set/St. Vincent University Hospital, Dublin [73] | 25 | 3-channel ECG | 6 months | 21 | 4 |
| Wisconsin Sleep Cohort [74,75] | 1545 | PSG, multiple sleep latency test | 1989–1993 | 1000 | 545 |
| STAGES -Stanford Technology Analytics and Genomics in Sleep [75] | 1500 | PSG | – | – | – |
| Apnea-ECG database [76] | 70 | ECG | 7–10 h | ||
| MIT-BIH [73] | 18 | PSG | 80 h | – | – |
| Sleep Heart Health Study (SHHS) [75,77] | 6441 | PSG | 1995–1998 | – | – |
Physiotherapy treatment for SA
| Author | Sample Size | Treatment | Improvements |
|---|---|---|---|
| Kumar, 2019 [38] | 23 | Yoga program for 3 months for mild-to-moderate SA and snoring issues | Positive impact on breathing pattern, oropharyngeal musculature, respiratory concerns. |
| Bankar, 2013 [39] | 2 groups | Two groups of SA patients formed; control group and yoga group | Yoga group scored higher quality of life PSQI value, reduced sleep disturbances, shorter sleep latency. |
| Khalsa, 2021 [40] | 2 groups | Two groups of insomnia patients formed: Kundalini yoga and sleep hygiene | Kundalini yoga group showed improved sleep quality |
| Kanchibhotla, 2021 [41] | 473 | Sudarshan Kriya yoga and breathing exercise | Positive impact on sleep quality. Improvement varied among patients based on age, gender, yoga practice frequency. |
| Kwiatkowska, 2008 [42] | – | Fuzzy logic-based treatment method to classify physical activities performed by OSA patients based on IPAQ | Improves monitoring of effectiveness of CPAP treatment, physical activities performed. |
| Khobarkar, 2022 [43] | – | Abhyanga Utsadana, Basti, oral medication with bitter herbs | Improvement shown in blood sugar level, BMI, waist-to-hip ratio, categories of BSQI. |
Recommender System for SA
| Author | Disease Prediction | Recommender System | ML/DL Algorithm | Data Set Used |
|---|---|---|---|---|
| Nanehkaran, 2022 [63] | Chronic Disease | collaborative filtering | K-NN classifier | PhysioNet data repository |
| Raza et al. 2023 [65] | SA | two-stage recommender system; precision = 89%, macro-average F1 score = 84% | – | MIMIC data set |
| Kaneriya et al. [68] | SA | Markov decision-based recommender system | Hidden Markov Model | – |
| Pinon et al., 2023 [66] | – | federated learning recommender system | – | historical disease-drug interactions and drug data |
| del Rio et al., 2023 [67] | Chronic Disease | – | restricted Boltzmann machine | wearable sensors connected on patient’s body |
| Casal-Guisande et al., 2023 [64] | OSA | personalized recommendation | ML classifier with fuzzy expert system | data set with 4400 patients from the Álvaro Cunqueiro Hospital (Vigo, Galicia, Spain) |
| Torres-Ruiz et al., 2023 [69] | COVID-19 | collaborative filtering | – | – |
| Chinyere et al., 2023 [70] | Hospital Recommendation | collaborative filtering | – | data collected through mobile/web application |
ML algorithms for SA diagnosis
| Author | Bio signal | ML Algorithm | Performance | Type of Classification | ||
|---|---|---|---|---|---|---|
| Accuracy (%) | Sensitivity | Specificity | ||||
| Sharma et al., 2023 [49] | EEG | K-NN, ensemble bagged trees (EbagT) | 92.85 | – | – | detection |
| Mencar et al., 2020 [50] | Questionnaire based data | SVM, RF, LR | 44.7 | – | – | prediction |
| Álvarez et al., 2020 [44] | SpO2, BP, HR | LR, SVM | 81.3 Kappa coefficient = 0.71 | – | – | AHI prediction |
| Shi et al., 2022 [28] | BP, SpO2 | GBM, XGBOOST | 88.5 | 0.713 | 0.873 | prediction, hypertension |
| Kristiensen et al., 2018 [51] | ECG | RF, KNN, SVM, ANN | 87.47 | – | – | classification |
| Pombo et al., 2020 [18] | ECG | SVM, LR | 82.12 | 0.8814 | 0.7229 | classification |
| Schrader et al., 2000 [52] | ECG, HRV | LDA | 88.31 | – | – | classification |
| Lin et al., 2006 [53] | ECG | DWT, ANN | – | 0.6964 | 0.4444 | classification |
| Xie & Minn, 2012 [54] | SpO2, ECG | KNN | 84.80 | – | – | prediction and classification |
Subject demographics
| Author | Data Sample Size (N) | Male (M) | Female (F) | Age Range/Mean Age | Time Frame | Type of Data |
|---|---|---|---|---|---|---|
| Peppard et al. [3] | 1520 | – | – | 37–85 | 2000–2015 | PSG |
| Han et al. [27] | 4014 | 2841 | 1173 | 53 | 2014–2021 | PSG, ESS questionnaire |
| Shi et al. [28] | 1493 | 1269 | 224 | – | 2019–2021 | PSG |
| Targa et al. [15] | 116 | 52 | 64 | 72–80 | 2015–2019 | PSG |
| Zarei et al. [29] | 25 | 21 | 04 | 28–68 | 2011 | PSG |
| Huttunen et al. [25] | 877 | 480 | 396 | 44–65 | 2015–2017 | PSG |
| Pombo et al. [18] | 70 | 57 | 13 | 27–63 | ECG |