Predicting effective continuous positive airway pressure in sleep apnea using an artificial neural network

Formulas to predict continuous positive airway pressure level using a home auto-adjusting device for obstructive sleep apnea treatment

PURPOSE

To develop equations to predict therapeutic continuous positive airway pressure (CPAPT) based on home-based CPAP titration, including the type of interface used.

METHOD

Retrospective study conducted in adult patients with obstructive sleep apnea (OSA) who used home-based autoCPAP titration (AutoSet S10, ResMed®). CPAPT was obtained manually through a visual analysis of autoCPAP data (CPAPV) and automatically using the 95th percentile pressure (CPAPP95). Multiple linear regression and K-fold cross-validation were applied. Independent variables were AHI, neck circumference (NC), BMI, and mask. Two formulas were generated based on mask and the Miljeteig and Hoffstein formula.

RESULTS

We included 702 patients (174 women), median age, BMI and AHI of 58 years, 32 kg/m2 and 32 ev/h, respectively. Predictors for CPAPv (M1) were BMI, NC, AHI and type of interface (R2: 0.19); and for CPAPP95 (M2), BMI, AHI and mask (R2: 0.09). Error and precision between the formulas and CPAPT were: 0 (CPAPV/CPAPP95), and - 3.2 to 3.2 (CPAPV) and - 4 to 4 cm H2O (CPAPP95). CPAPV was higher with oronasal mask (10 vs. 9 cm H2O, p < 0.01). Accuracy defined as; a difference ± 2 cm H2O between estimated CPAP and CPAPT was greater in M1 than in M2 (79% vs. 64%, p < 0.01).

CONCLUSION

In both models, calculated error was close to zero. CPAPV (± 3.2 cm H2O) showed more precision than CPAPP95 (± 4 cm H2O). With M1 (CPAPV), 79% of patients could start CPAP with reasonable accuracy (error of ± 2 cm H2O).

Application of artificial intelligence in the diagnosis of sleep apnea

STUDY OBJECTIVES

Machine learning (ML) models have been employed in the setting of sleep disorders. This review aims to summarize the existing data about the role of ML techniques in the diagnosis, classification, and treatment of sleep-related breathing disorders.

METHODS

A systematic search in Medline, EMBASE, and Cochrane databases through January 2022 was performed.

RESULTS

Our search strategy revealed 132 studies that were included in the systematic review. Existing data show that ML models have been successfully used for diagnostic purposes. Specifically, ML models showed good performance in diagnosing sleep apnea using easily obtained features from the electrocardiogram, pulse oximetry, and sound signals. Similarly, ML showed good performance for the classification of sleep apnea into obstructive and central categories, as well as predicting apnea severity. Existing data show promising results for the ML-based guided treatment of sleep apnea. Specifically, the prediction of outcomes following surgical treatment and optimization of continuous positive airway pressure therapy can be guided by ML models.

CONCLUSIONS

The adoption and implementation of ML in the field of sleep-related breathing disorders is promising. Advancements in wearable sensor technology and ML models can help clinicians predict, diagnose, and classify sleep apnea more accurately and efficiently.

CITATION

Bazoukis G, Bollepalli SC, Chung CT, et al. Application of artificial intelligence in the diagnosis of sleep apnea. J Clin Sleep Med. 2023;19(7):1337-1363.

Predictive performances of 6 data mining techniques for obstructive sleep apnea-hypopnea syndrome

PURPOSE

This study compared the effects of 6 types of obstructive sleep apnea-hypopnea syndrome (OSAHS) prediction models to develop a reference for selecting OSAHS data mining tools in clinical practice.

METHODS

This cross-sectional study included 401 cases. They were randomly divided into 2 groups: training (70%) and testing (30%). Logistic regression, a Bayesian network, an artificial neural network, a support vector learning machine, C5.0, and a classification and regression tree were each adopted to establish 6 prediction models. After training, the 6 models were used to test the remaining samples and calculate the correct and error rates of each model.

RESULTS

Twenty-one input variables for which the difference between the patient and nonpatient groups was statistically significant were considered. The models found the abdominal circumference, neck circumference, and nocturia ≥2 per night to be the most important variables. The support vector machine, neural network, and C5.0 models performed better than the classification and regression tree, Bayesian network, and logistic regression models.

CONCLUSIONS

In terms of predicting the risk of OSAHS, the support vector machine, neural network, and C5.0 were superior to the classification and regression tree, Bayesian network, and logistic regression models. However, such results were obtained based on the data of a single center, so they need to be further validated by other institutions.

A predictive model for optimal continuous positive airway pressure in the treatment of pure moderate to severe obstructive sleep apnea in China

Background

Numerous predictive formulas based on different ethnics have been developed to determine continuous positive airway pressure (CPAP) for patients with obstructive sleep apnea (OSA) without laboratory-based manual titrations. However, few studies have focused on patients with OSA in China. Therefore, this study aimed to develop a predictive equation for determining the optimal value of CPAP for patients with OSA in China.

Methods

526 pure moderate to severe OSA patients with attended CPAP titrations during overnight polysomnogram were spited into either formula derivation (419 patients) or validation (107 patients) group according to the treatment time. Predictive model was created in the derivation group, and the accuracy of the model was tested in the validation group.

Results

Apnea hypopnea index (AHI), body mass index (BMI), longest apnea time (LAT), and minimum percutaneous oxygen saturation (minSpO₂) were considered as independent predictors of optimal CPAP through correlation analysis and multiple stepwise regression analysis. The best equation to predict the optimal value of CPAP was: CPAPpred = 7.581 + 0.020*AHI + 0.101*BMI + 0.015*LAT-0.028*minSpO₂ (R² = 27.2%, p < 0.05).The correlation between predictive CPAP and laboratory-determined manual optimal CPAP was significant in the validation group (r = 0.706, p = 0.000). And the pressure determined by the predictive formula did not significantly differ from the manually titrated pressure in the validation cohort (10 ± 1 cmH₂O vs. 11 ± 3 cmH₂O, p = 0.766).

Conclusions

The predictive formula based on AHI, BMI, LAT, and minSpO₂ is useful in calculating the effective CPAP for patients with pure moderate to severe OSA in China to some extent.

Using craniofacial characteristics to predict optimum airway pressure in obstructive sleep apnea treatment

Introduction

Manual titration is the gold standard to determinate optimal continuous positive airway pressure, and the prediction of the optimal pressure is important to avoid delays in prescribing a continuous positive airway pressure treatment.

Objective

To verify whether anthropometric, polysomnographic, cephalometric, and upper airway clinical assessments can predict the optimal continuous positive airway pressure setting for obstructive sleep apnea patients.

Methods

Fifty men between 25 and 65 years, with body mass indexes of less than or equal to 35 kg/m² were selected. All patients had baseline polysomnography followed by cephalometric and otolaryngological clinical assessments. On a second night, titration polysomnography was carried out to establish the optimal pressure.

Results

The average age of the patients was 43 ± 12.3 years, with a mean body mass index of 27.1 ± 3.4 kg/m² and an apnea–hypopnea index of 17.8 ± 10.5 events per hour. Smaller mandibular length (p = 0.03), smaller atlas–jaw distance (p = 0.03), and the presence of a Mallampati III and IV (p = 0.02) were predictors for higher continuous positive airway pressure. The formula for the optimal continuous positive airway pressure was: 17.244 − (0.133 × jaw length) + (0.969 × Mallampati III and IV classification) − (0.926 × atlas–jaw distance).

Conclusion

In a sample of male patients with mild-to-moderate obstructive sleep apnea, the optimal continuous positive airway pressure was predicted using the mandibular length, atlas–jaw distance and Mallampati classification.

A new predictive model for continuous positive airway pressure in the treatment of obstructive sleep apnea

BACKGROUND

Numerous mathematical formulas have been developed to determine continuous positive airway pressure (CPAP) without an in-laboratory titration study. Recent studies have shown that style of CPAP mask can affect the optimal pressure requirement. However, none of the current models take mask style into account. Therefore, the goal of this study was to develop new predictive models of CPAP that take into account the style of mask interface.

METHODS

Data from 200 subjects with attended CPAP titrations during overnight polysomnograms using nasal masks and 132 subjects using oronasal masks were randomized and split into either a model development or validation group. Predictive models were then created in each model development group and the accuracy of the models was then tested in the model validation groups.

RESULTS

The correlation between our new oronasal model and laboratory determined optimal CPAP was significant, r = 0.61, p < 0.001. Our nasal formula was also significantly related to laboratory determined optimal CPAP, r = 0.35, p < 0.001. The oronasal model created in our study significantly outperformed the original CPAP predictive model developed by Miljeteig and Hoffstein, z = 1.99, p < 0.05. The predictive performance of our new nasal model did not differ significantly from Miljeteig and Hoffstein's original model, z = -0.16, p < 0.90. The best predictors for the nasal mask group were AHI, lowest SaO2, and neck size, whereas the top predictors in the oronasal group were AHI and lowest SaO2.

CONCLUSION

Our data show that predictive models of CPAP that take into account mask style can significantly improve the formula's accuracy. Most of the past models likely focused on model development with nasal masks (mask style used for model development was not typically reported in previous investigations) and are not well suited for patients using an oronasal interface. Our new oronasal CPAP prediction equation produced significantly improved performance compared to the well-known Miljeteig and Hoffstein formula in patients titrated on CPAP with an oronasal mask and was also significantly related to laboratory determined optimal CPAP.

Mathematical Equations to Predict Positive Airway Pressures for Obstructive Sleep Apnea: A Systematic Review

Objective. To systematically review the international literature for mathematical equations used to predict effective pressures for positive airway pressure (PAP) devices. Methods. Google Scholar, PubMed, Scopus, Embase, Web of Science, CINAHL, and The Cochrane Library were searched through June 27, 2015. The PRISMA statement was followed. There was no language limitation. Results. 709 articles were screened, fifty were downloaded, and twenty-six studies presented equations that met the inclusion and exclusion criteria. In total, there were 4,436 patients in the development phases and 3,489 patients in the validation phases. Studies performed multiple linear regressions analyses as part of the equation(s) development and included the following variables: physical characteristics, polysomnography data, behavioral characteristics, and miscellaneous characteristics, which were all predictive to a variable extent. Of the published variables, body mass index (BMI) and mean oxygen saturation are the most heavily weighted, while BMI (eighteen studies), apnea-hypopnea index (seventeen studies), and neck circumference (eleven studies) were the variables most frequently used in the mathematical equations. Ten studies were from Asian countries and sixteen were from non-Asian countries. Conclusion. This systematic review identified twenty-six unique studies reporting mathematical equations which are summarized. Overall, BMI and mean oxygen saturation are the most heavily weighted.

Development of a system for telemonitoring of respiration parameters for patients with Obstructive Sleep Apnea

Obstructive Sleep Apnea is a chronic sleep disorder affecting a large number of the global population. Telemonitoring has been successfully evaluated as an alternative method to traditional care. This paper identifies drawbacks of the current telemonitoring approaches and presents a universal wireless system for continuous monitoring of basic respiration parameters. The proposed system monitors four parameters, namely respiratory flow, airway pressure, Carbon Dioxide (CO2) and Oxygen (O2) gas concentrations. Data are wirelessly transmitted to a computer which acts as a web server. The system will allow remote evaluation of home ventilation support efficiency and the application of custom algorithms for decision support and respiration event detection.

Determination of new prediction formula for nasal continuous positive airway pressure in Turkish patients with obstructive sleep apnea syndrome

BACKGROUND

Race/ethnicity may play an important role in determining body size, severity of obstructive sleep apnea syndrome (OSAS), and effective continuous positive airway pressure (CPAP) (Peff). Turkey is composed of different ethnic groups. Therefore, the aims of this study were to determine new prediction formula for CPAP (Ppred) in Turkish OSAS patients, validate performance of this formula, and compare with Caucasian and Asian formulas.

METHODS

Peff of 250 newly diagnosed moderate-to-severe OSAS patients were calculated by in-laboratory manual titration. Correlation and multiple linear regression analysis were used to model effects of ten anthropometric and polysomnographic variables such as neck circumference (NC) and oxygen desaturation index (ODI) on Peff. New formula was validated in different 130 OSAS patients and compared with previous formulas.

RESULTS

The final prediction formula was [Formula: see text]. When Peff of control group was assessed, it was observed that mean Peff was 8.39 ± 2.00 cmH(2)O and Ppred was 8.23 ± 1.22 cmH(2)O. Ppred was within ±3 cmH(2)O of Peff in 96.2% patients. Besides, Peff was significantly correlated with new formula, and prediction formulas developed for Caucasian and Asian populations (r = 0.651, p < 0.001, r = 0.648, p < 0.001, and r = 0.622, p < 0.001, respectively).

CONCLUSIONS

It is shown that level of CPAP can be successfully predicted from our prediction formula, using NC and ODI and validated in Turkish OSAS patients. New equation correlates with other formulas developed for Caucasian and Asian populations. Our simple formula including ODI, marker of intermittent hypoxia, may be used easily in different populations.

Utility of formulas predicting the optimal nasal continuous positive airway pressure in a Greek population

BACKGROUND

There have been reports that optimal CPAP pressure can be predicted from a previously derived formula, with the Hoffstein formula being the most accurate and accepted in the literature so far. However, the validation of this predictive model has not been applied in different clinical settings. Our aim was to compare both the Hoffstein prediction formula and a newly derived formula to the CPAP pressure setting assessed during a formal CPAP titration study.

METHODS

We prospectively studied 1,111 patients (871 males/240 females) with obstructive sleep apnea hypopnea syndrome (OSAHS) undergoing a CPAP titration procedure. In this large population sample, we tested the Hoffstein formula, utilizing body mass index (BMI), neck circumference and apnea/hypopnea index (AHI), and we compared it with our new formula that included not only AHI and BMI but also smoking history and gender adjustment.

RESULTS

We found that using the Hoffstein prediction formula, successful prediction (predicted CPAP pressure within ±2 cm H(2)O compared to the finally assessed optimum CPAP pressure during titration) was accomplished in 873 patients (79%), with significant correlation between CPAP predicted pressure (CPAPpred(1)) and the optimum CPAP pressure (CPAPopt) [r = 0.364, p < 0.001]. With the new formula, including smoking history and gender adjustment, successful prediction was accomplished in 1,057 patients (95%), with significant correlation between CPAP predicted pressure (CPAPpred(2)) and the CPAPopt (r = 0.392, p < 0.001). However, there was a highly significant correlation between the two formulas (r = 0.918, p < 0.001).

CONCLUSIONS

We conclude that the level of CPAP necessary to abolish sleep apnea can be successfully predicted from both equations, using common clinical measurements and prediction formulas that may be useful in calculating the starting pressure for initiating CPAP titration. It may also be possible to shorten CPAP titration and perhaps in selected cases to combine it with the initial diagnostic study.

Predicting optimal CPAP by neural network reduces titration failure: a randomized study

PURPOSE

Continuous positive airway pressure (CPAP) is considered the standard therapy for obstructive sleep apnea syndrome. In the absence of standard protocol, CPAP titration may be unsuccessful. The purpose of this study was to test the hypothesis that application of an artificial neural network (ANN) to CPAP titration would achieve an optimal CPAP pressure within a shorter time interval and would lead to a decrease in CPAP titration failure.

METHODS

One hundred fifteen patients were randomized 1:1 to either conventional CPAP titration (n = 58) or to an ANN-guided CPAP titration (n = 57). Both groups were assessed for time to optimal CPAP pressure, for titration failure, and for CPAP compliance therapy.

RESULTS

Patients in the ANN-guided CPAP titration arm were able to achieve optimal CPAP at a shorter time interval compared to the conventional group (198.7 +/- 143.8 min versus 284.0 +/- 126.5 min) (p < 0.001). There was also a lower titration failure in patients randomized to the ANN-guided CPAP titration arm (16%) compared to the conventional arm (36%) (p = 0.02). Compliance with treatment did not differ across the two arms.

CONCLUSIONS

The use of ANN for guiding CPAP titration may be superior to the conventional method in maximizing the time to achieve optimal CPAP and in reducing CPAP titration failure.