The oxygen desaturation measured by optical sensor at fingertip but not at the arm is diminished by continuous positive airway pressure in obstructive sleep apnea

: Obstructive sleep apnea (OSA) is a common sleep disorder, and continuous positive airways pressure (CPAP) is the most effective treatment. Poor adherence is one of the major challenges in CPAP therapy. The recent boom of wearable optical sensors measuring oxygen saturation makes the at-home multiple-night CPAP titrations possible, which may essentially improve the adherence of CPAP therapy by optimizing its pressure in a real-life setting economically. We tested whether the oxygen desaturations (OD) measured in the arm muscle (arm_OD) by gold-standard frequency-domain multi-distance near-infrared spectroscopy (FDMD-NIRS) changes with titrated CPAP pressures in OSA patients together with polysomnography. We found that the arm_OD (2.08 ± 1.23%, mean ± standard deviation) was significantly smaller (P-value <0.0001) than the fingertip OD (finger_OD) (4.46 ± 2.37%) measured by polysomnography pulse oximeter. Linear mixed-effects models suggested that CPAP pressure was a significant predictor for finger_OD but not for arm_OD. Since FDMD-NIRS measures a mixture of arterial and venous OD, whereas fingertip pulse oximeter measures arterial OD, our results of no association between arm_OD and finger_OD indicate that the arm_OD mainly represented venous desaturation. Arm_OD measured by near-infrared optical sensors may be not a suitable indicator of the effectiveness of CPAP titration.


Introduction
Obstructive sleep apnea (OSA) is the most prevalent respiratory sleep disorder occurring in 9-38% of the general population [1], and it is a high risk factor for cardio-/cerebrovascular diseases [2,3]. Continuous positive airways pressure (CPAP) or automatic positive airway pressure (APAP) therapy is currently the most effective treatment for OSA [4,5]. However, there are still many challenges in OSA diagnosis and treatment, particularly simple low-cost new diagnostic technologies that can be easily used by the patients for home recordings, are urgently needed [6,7]. This urgency is due to observations that 1) the majority of suspected OSA patients remain undiagnosed [8,9] in many countries due to the lack of sleep laboratories/specialists and the high-cost of in-lab polysomnography (PSG) diagnosis, and 2) long-term home recordings employing these easy-to-use technologies will be helpful to longitudinally monitor the CPAP treatment effects and adherence or even to select patients who will benefit most from treatment (i.e., precision or personalized medicine) [6,10].
Recently, low-cost wearable devices, such as smartwatches (e.g., the new products from Fitbit [11], Garmin [12], Huami [13], Huawei [14]) and armbands (e.g., Humon [15,16], Moxy [17,18], Artinis [19] and Biofourmis [20]) have implemented the function of measuring peripheral capillary (SpO2) or muscle tissue (StO2) oxygen saturation at the arm or wrist using near-infrared light, making them possible alternatives for at-home OSA measurement. Unlike the in-lab PSG fingertip pulse oximetry, in which the detector measures the light transmitted through the fingertip (i.e., the transmission photoplethysmography, T-PPG), these wearable devices assess the SpO2 or StO2 by measuring the changes in the backscattering light from the wrist or arm based on the modified Beer-Lambert law (MBLL) [21][22][23][24]. That is, their light sources that are usually two or more near-infrared light wavelengths mainly absorbed by oxygenated hemoglobin (HbO2) and deoxygenated hemoglobin (HHb) in biological tissues, and detectors are placed on the same side of the measured tissues, because the wrist and arm are too thick to be penetrated through by the light. While the wrist wearable devices mainly use reflectance photoplethysmography (R-PPG) or pulse oximetry [24][25][26], the armbands are mainly based on simple continuous wave near-infrared spectroscopy (CW-NIRS) methods [16,18].
Measuring OSA by these low-cost wearable devices is unfortunately still not viable, because none of them has been licensed or certified as a medical diagnostic device by the United States Food and Drug Administration or CE marking [27]. In our recent study [28] we compared event-by-event oxygen desaturation (OD) at the fingertip (finger_OD) measured by gold-standard in-lab PSG fingertip T-PPG with the OD at arm (arm_OD) measured by gold-standard frequency-domain multi-distance near-infrared spectroscopy (FDMD-NIRS) in OSA patients during naïve sleep and during CPAP titrations. Our results of Bland-Altman plots gave poor agreements between finger_OD and arm_OD in sleep hypopneas, suggesting that the absolute value of arm_OD is not suitable to define sleep hypopneas according to the standard clinical criterion of ≥ 3% OD drop. It could lead to large false negative results in measuring OSA events, thus underestimating the Apnea-Hypopnea index (AHI) compared to in-lab PSG.
However, whether the qualitative changes in arm_OD can indicate the CPAP titration effectiveness remains unknown. The CPAP/APAP machine itself can measure AHI, although its accuracy has been criticized [29][30][31]. Two major challenges in CPAP/APAP treatment are poor adherence to treatment [32] and persisting hypoxemia and daytime syndromes despite therapy in some patients, for whom oxygen supplementation may be needed [6]. Logically, 'A does not equal to B' does not necessary mean that 'changes in A is not associated with changes in B'. For example, obesity is a risk of OSA. Large BMI cannot be used to replace large AHI to define OSA, but the decrease in BMI is associated with decrease in AHI [33]. If the changes in arm_OD can be associated with changes in CPAP/APAP pressures, then measuring changes in arm_OD with a wearable device at home could be helpful to better understand those patients, e.g., to identify subgroups with/without improvement in OD when CPAP/APAP pressure changes. It could also make at-home multiple night CPAP titrations possible, i.e., to find the optimal CPAP pressure that can at best restore the respiratory events and arm_OD, which may essentially improve the adherence of CPAP therapy by optimizing its pressures in a real-life setting economically.
Therefore, in this study we quantify the dynamic changes in arm_OD in OSA events under titrated CPAP pressures using regression analysis. We use the same dataset that has been recently published in [28], considering its major advantages, which include: 1) the CPAP pressures were well-controlled stepwise and increased hourly during the titrations; 2) the simultaneously measured finger_OD can serve as a control for CPAP titration effectiveness, i.e., an effective titration procedure should restore stepwise the fingertip SpO2 desaturation; 3) the comparisons between finger_OD (i.e., arterial desaturation) and arm_OD (i.e., the mixture of arterial and venous desaturation [34,35]) may provide new insights into the peripheral desaturation in response to OSA events during CPAP therapy. We hypothesized that the changes in arm_OD may be associated with changes in CPAP pressures, considering that our stepwise incremental CPAP titration gradually opened the upper airway to increase oxygen supply.

Materials and Methods
Thirty newly diagnosed OSA patients (age [mean ± standard deviation, SD]: 54. Medicine manual [36]. The discrepancy between these two technologists was resolved by discussion or recommendation by an experienced neurophysiologist. have been well validated in different physical blood-lipid models [38,40,41] and in vivo studies [42][43][44][45]. It has been used as a gold-standard reference measurement of StO2 for the validations or calibrations of wearable CW-NIRS armbands [16], and portable CW-NIRS oximeters including those have received FDA clearance [41,46]. Its light emitters, four laser diodes at 690-nm wavelength and four laser diodes at 830-nm wavelength are coupled into four light sources and are high frequency modulated at 110 MHz. The light can penetrate into the measured tissues with a depth of several centimeters when the four light sources are aligned and placed at 2 cm, 2.5 cm, 3 cm and 3.5 cm from an optical fiber bundle connected to the photomultiplier tube detector. The sampling rate of FDMD-NIRS recording was set as 5.2 Hz. The Imagent system was calibrated on an optical phantom block to exclude the uncertainty of measurements due to machine errors before each recording. The raw measured NIRS data were subjected to a low-pass (<0.08 Hz) zero-phase filter designed using Hanning window to remove the physiological noises including heart rate, respiratory noise and spontaneous slow hemodynamic oscillations [28,47,48]. The filtered data were then smoothed using the robust locally weighted scatter plot smoothing method [47,49].
The data analysis procedure was shown in Fig.1. After a standard PSG scoring, per-hour AHI under each CPAP pressure was calculated, i.e., the number of event was divided by the sleep time under each CPAP pressure per-hour in the titration protocol.
Obstructive apneas (n=29) and hypopneas (n=31) were excluded from analysis if their SpO2 desaturations were larger than 15% to exclude outliers and potentially unreliable measurements caused by instrument errors [26,28]. In each patient all the events under a specific CPAP pressure were also excluded if the corresponding sleep duration under that pressure was shorter than 20 minutes, to exclude the unreliable calculation of the per-hour AHI, e.g., if the sleep duration under some CPAP pressure was just a few minutes while the patient had a number of apneas/hypopneas, the calculated per-hour AHI can be extremely large but its value was unreliable because of short sleep time.
Then linear mixed-effects model (LMM) with a random intercept by patients was used to predict the arm_OD and finger_OD caused by the respiratory events, respectively.
Explanatory variables were demographic variables (i.e., age, sex, BMI, AHI of the diagnostic night measured by PSG) and parameters that can be measured by CPAP machine or wearable devices, i.e., types of respiratory events, durations of event, sleep stages, mean HR during the events, per-hour AHI under each pressure, and CPAP pressures. Stepwise regression using backward elimination was performed to automatically select the best predictors. Then the final LMM models were built using these best predictors. We reported both the conditional R 2 [50] and Ω 2 [51] to assess the goodness of fit of our final selected models. The LMM models were done using the R package lme4 (function lmer) and stepwise regressions were done using the R package lmerTest (function step).

Results
In total, 505 obstructive apneas and 2185 hypopneas events were analyzed. The median of the number of events acquired from our patients was 75 with IQR between 59 and 110. Fig.2 illustrates typical changes in fingertip SpO2 and arm StO2 desaturations in OSA events. ODs triggered by sleep apneas occurred in both SpO2 and StO2, although the baseline StO2 values (mostly between 60-70%) were smaller than SpO2 (mostly above 90%) because StO2 were from both venous and arterial blood. The mean arm_OD (2.08 ± 1.23%) was significantly smaller (paired t-test, P-value<0.0001) than the mean finger_OD (4.46 ± 2.37%). There was no correlation between the degrees of arm_OD and finger_OD, indicated by the Pearson's correlation coefficient of 0.08 (P-value <0.0001).   It could be possible that the normalized changes rather than the raw values of arm_OD were associated with CPAP pressures, since the StO2 baseline level (68.6 ± 6.4%) before the desaturations was obviously much smaller than that of SpO2 (usually above 90%). We therefore normalized the arm_OD to its baseline and repeated the LMM analysis. The CPAP pressure was still automatically excluded from the final selected model in stepwise regression, in which the final selected predictors were the same as shown in Table 1, i.e., the duration of events (estimated coefficient 0.033, P-value <0.0001), mean HR within events (estimated coefficient -0.03, P-value <0.0001), per-hour AHI (estimated coefficient 0.0028, P-value = 0.00058).

Discussion
In this study we test whether oxygen desaturations measured in the arm muscle change with CPAP pressures during CPAP titration in patients with OSA. In contrast to our recent study testing the agreement between arm and fingertip oxygen desaturations in sleep hypopneas using Bland-Altman plots [28], here we use linear regression (i.e., LMM) to study the association between CPAP pressures and oxygen desaturations at arm and fingertip. Contrary to our hypothesis, we only found association between CPAP pressures and the oxygen desaturations at the fingertip but not in the arm muscle.
Our negative results suggest that muscular oxygen desaturation may be not a suitable indicator of the effectiveness of CPAP titration. Thus, the usefulness of wearable devices measuring arm StO2 in CPAP therapy is questionable.
Only fingertip SpO2 but not the arm StO2 reflects the reduction of desaturations during CPAP titration, probably because venous blood contributing to StO2 reduces the sensitivity of StO2 in response to CPAP pressures compared to SpO2. The NIRS StO2 is the proportion of HbO2 in the measured biological tissues including arterial, capillary and venous compartments. It can be expressed as:

StO2=a×SaO2+b×SvO2
(1) where SaO2 and SvO2 are the arterial and venous oxygen saturation [34,35]. SaO2 is approximately equal to SpO2 as they are both arterial oxygen saturation (usually close to 100%). SvO2 is usually 65-75% [52]. The ratio of coefficients a/b is the arterial-to-venous volume ratio (AVR), and a+b=1. Most commercially available NIRS oximeters including the FDA-certificated medical devices usually fix the AVR as either 25%/75% or 30%/70% [53][54][55][56][57][58][59][60][61][62][63] but never validate them in OSA. If we assume that the fixed AVR model was valid in OSA, we could expect that arm_OD highly correlates with finger_OD considering their mathematical relationship, e.g., 1% decrease in fingertip arterial SpO2 may correspond to 0.25% or 0.3% decrease in arm StO2 because only a percent (25% or 30%) of the 1% SpO2 desaturation can contribute to StO2 desaturation according to Formula (1). Thus, similar as the results of finger_OD in response to CPAP titration shown in Table 2, arm_OD should also decrease stepwise with increasing CPAP pressures. However, the lack of correlation between the changes in finger_OD and arm_OD and the missing association between CPAP pressures and arm_OD contradict this assumption, suggesting the fixed AVR model is unlikely to be valid in OSA. In fact OSA event is actually associated with increased vasoconstriction in peripheral limb arteries and arterioles [64,65], which suggests the AVR in the arm muscle is hardly constant during OSA. Most likely the coefficient a decreases while coefficient b increases in OSA because: 1) a+b=1 thus an increase in one coefficient must be associated with a decrease in the other one, and arterial vessels have stronger capacity to constrict than venous vessels; 2) it is known that blood pressure, HR, left ventricular stroke volume and cardiac preload all decrease during apnea/hypopnea events [47,64,[66][67][68], indicating more blood may be held in the venous vascular bed. Therefore, the contribution of venous blood to StO2 increases while arterial blood contribution decreases during OSA events, leading to a reduction in the sensitivity of StO2 in response to SpO2 changes.
This interpretation also fits our results that arm_OD has a negative association with the mean HR during the events (Table 1), i.e., higher HR may indicate less vasoconstriction (i.e., relative larger a) and relatively more arterial blood supply to the muscle tissues, ergo arm_OD is smaller.
To the best of our knowledge, NIRS has not been used to measure the peripheral hemodynamics (i.e., muscular hemodynamics) in OSA. Our results of no correlation between finger_OD and arm_OD and the aforementioned increased venous contribution to StO2 during OSA events indicate that our arm_OD measured by NIRS is likely to mainly represent the OD in venous blood. Our results (Table 1 and Table 2) thus provide new insights into the changes in peripheral oxygen desaturation in OSA events during CPAP, that 1) the arterial but not venous desaturation is more sensitive to changes in Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 2 November 2021 CPAP pressure; 2) arterial but not venous desaturation depends on the types of events (i.e., apnea causes larger desaturation than hypopnea) and sleep stages (light sleep causes larger desaturation than others); 3) longer events cause larger desaturations in both arterial and venous blood as indicated by longer hypoxia that cause stronger oxygen extraction from both arterial and venous vascular bed.
In our recent study using the same database we reported poor agreements (analyzed by Bland-Altman plots) between arm_OD and finger_OD in sleep hypopneas, and thus the reliability of AHI measured by StO2 desaturation using wearable or portable optical sensors based on NIRS technique is questionable [28]. This conclusion could be extended to wearable optical sensors based on R-PPG technique like wrist smartwatches, because they face the same problem of venous blood influence as NIRS [28]. A recent study tested the accuracy of wrist R-PPG smartwatch in measuring SpO2, when SaO2 measured from blood samples with a co-oximeter changed from 100% to 70% [69]. In that study, the Bland-Altman plot gave a broad 95% lower (i.e., approximate -4%) and upper (i.e., approximate 6-7%) limits of agreement between smartwatch and co-oximeter measurements, which are similar as those reported in our study [28].

Conclusions
Although the recent boom of wearable optical sensors like smartwatches and armbands offers a possibility of assessing OSA and multiple-night CPAP titrations at home, our negative results should warn the general public and sleep researcher/clinicians to be cautious with these wearable devices until those products are clinically and experimentally validated. Probably more sophisticated algrithms such as machine learning are needed to derive some peripheral parameters than can correctly measure SpO2 using wearable devices. Our results also suggest that the muscular StO2 desaturation measured by NIRS may mainly represent venous desaturation in OSA. We suggest that more studies including the gold-standard invasive measurements of SaO2 Institutional Review Board Statement: The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the local ethical commission of Northwest Switzerland.

Informed Consent Statement:
Written informed consent has been obtained from the patient(s) to publish this paper . Data Availability Statement: The raw data supporting the conclusions of this article are available from the corresponding author on reasonable request.