Method of detecting and tracking respiratory control instability and expressed high loop gain by self-similarity analysis

Abstract

Disclosed herein are example methods of assessing sleep disorders by detecting and tracking respiratory control instability by self-similarity analysis. A method of assessing sleep disorders of a patient includes receiving signals from the patient over a period of time while the patient is asleep, identifying a complex within the received signals, determining a ratio of a maximum envelope height to a minimum envelope height within the complex, determining a duration between a maximum envelope height of the complex and a subsequent maximum envelope of a subsequent complex, computing a convolution of a positive envelope and a rotated negative envelope of the complex, assessing a time location of a maximum distance between the positive envelope and the negative envelope, and determining a self-similarity of the received signals.

Description

METHOD OF DETECTING AND TRACKING RESPIRATORY CONTROL INSTABILITY AND EXPRESSED HIGH LOOP GAIN BY SELF- SIMILARITY ANALYSISCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional Application Number 63 / 558,429, filed on February 27, 2024, which is hereby incorporated by reference in its entirety.STATEMENT REGARDING FEDERALLY-SPONSORED RESEARCH AND DEVELOPMENT

[0002] This invention was made with government support under grant NS 120947 awarded by the National Institutes of Health. The government has certain rights in the invention.BACKGROUND OF THE INVENTIONField of the Invention

[0003] Embodiments herein relate to a method for detecting respiratory control instability and expressed high loop gain.Background

[0004] Obstructive sleep apnea (OSA) is a condition that can increase the risk of cardiometabolic and neurodegenerative diseases (1, 2). Several driver endotypes are now recognized to be important in the pathophysiology of OSA, resulting in an admixture of anatomical and non-anatomical mechanisms (3, 4). Of the latter, high loop gain (HLG) is considered a key treatment target for precision sleep apnea care. However, recognition of LG driving sleep apnea outside classic central apnea of Hunter-Cheyne- Stokes Respiration has proven difficult. Several methods have been proposed, including hypoxic- hypercapnic gas administration (5-9), proportional assist ventilation (10, 11), breathholding estimates (12-14), mathematical modeling of underlying breathing mechanics(15-17), cardiopulmonary coupling (18, 19), and mathematical estimates of selfsimilarity (SS) (20).

[0005] Multiple mechanisms are involved in the pathogenesis of obstructive sleep apnea (OSA). Elevated loop gain is a key target for precision OSA care and is associated with treatment intolerance when the upper airway is the sole therapeutic target. Morphological or computational estimation of LGis not yet widely available or fully validated - there is a need for improved phenotyping / endotyping of apnea to advance its therapy and prognosis.

[0006] While the non-rapid eye movement sleep carbon dioxide (CO2) reserve can improve with sleep apnea treatment (21, 22), therapies targeting upper airway obstruction alone may result in adverse physiological and clinical outcomes when HLG is present (23-27). One-third to half of patients receiving CPAP continue to experience residual events (28-30). Loop gain can be lowered by supplemental oxygen (31) or controlling the inspiratory CO2 level (32), and by medications, e.g., acetazolamide, zonisamide or sulthiame (33-36). To help guide precision sleep apnea treatment strategies there is a need for practically applicable and clinically predictive phenotyping / endotyping of apnea.SUMMARY OF THE INVENTION

[0007] Described herein are example methods and systems for a new algorithm to assess self-similarity (SS) as a signature of elevated loop gain using respiratory effort signals, and presents an algorithm that uses SS to predict the probability of failure (high residual event counts) of continuous positive airway pressure (CPAP) therapy.

[0008] In the embodiments presented herein, a method of assessing sleep disorders of a patient is described. The method includes receiving signals from the patient over a period of time while the patient is asleep, the signals being representative of sleep respiratory effort / motion of the patient, identifying a complex within the received signals, the identifying including ascertaining an envelope of the received signals over the period of time, determining a ratio of a maximum envelope height to a minimum envelope height within the complex, determining a duration between a maximum envelope height of the complex and a subsequent maximum envelope of a subsequent complex, computing a convolution of a positive envelope and a rotated negative envelope of the complex, the rotated negative envelope being a rotation of the negative envelope about a time axis,assessing a time location of a maximum distance between the positive envelope and the negative envelope, and determining a self-similarity of the received signals based on the identifying, determining, determining, computing and assessing.

[0009] Further features and advantages, as well as the structure and operation of various embodiments, are described in detail below with reference to the accompanying drawings. It is noted that the specific embodiments described herein are not intended to be limiting. Such embodiments are presented herein for illustrative purposes only. Additional embodiments will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein.BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate the present disclosure and, together with the description, further serve to explain the principles of the present disclosure and to enable a person skilled in the relevant art(s) to make and use the present disclosure. The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.

[0011] FIG. 1 illustrates an example diagram of a sensor device placed on a patient, according to embodiments of the present disclosure.

[0012] FIGS. 2A-2F illustrate comparisons of regular breathing and self-similar breathing according to embodiments of the present disclosure, wherein FIGS. 2A-2C illustrate regular breathing and FIGS. 2D-2F illustrate self-similar breathing.

[0013] FIG. 3 illustrates a method of processing breathing oscillations according to embodiments of the present disclosure, by (1) computing a convolution between the positive and the rotated negative envelope, and (2) assessing the location of the maximum distance between the positive and negative envelope.

[0014] FIG. 4 illustrates an example of computing self-similarity in breathing using convolution scores across neighboring envelopes of oscillations.

[0015] FIGS. 5A-5F illustrate effort trace examples comparing example algorithm and self-similarity algorithm according to embodiments of the present disclosure.

[0016] FIGS. 6A-6D illustrates example traces of patients according to embodiments of the present disclosure.

[0017] FIG. 7 illustrates an exemplary chart of high loop gain breathing oscillations per sleep stage according to embodiments of the present disclosure.

[0018] FIGS. 8A-8C illustrate Receiver Operating Characteristic (ROC), Precision- Recall (PR), and calibration curves of the central apnea index (CAI), hypoxic burden (Burden), and self-similarity (SS) as individual predictive features for CPAP failure, according to embodiments of the present disclosure.

[0019] FIGS. 9 A and 9B illustrate overnight recordings from patients exhibiting mild and severe SS, respectively, according to embodiments of the present disclosure

[0020] FIGS. 10A-10F illustrate receiver Operating Characteristic (ROC), Precision- Recall (PR), and calibration curves of the central apnea index (CAI), the hypoxic burden (Burden), and self-similarity (SS) for all patients with a pre-treatment, according to embodiments of the present disclosure.

[0021] FIGS. 11 A-l IF illustrate Receiver Operating Characteristic (ROC), Precision- Recall (PR), and calibration curves of the central apnea index (CAI), the hypoxic burden (Burden), and self-similarity (SS) for all patients with a pre-treatment, according to embodiments of the present disclosure.

[0022] FIGS. 12A and 12B illustrate heat-maps showing the relationship between the level of self-similarity and the hypoxic burden, according to embodiments of the present disclosure.

[0023] FIG. 13 illustrates flowchart depicting a method for assessing sleep disorders of a patient, according to embodiments of the present disclosure.

[0024] FIG. 14 illustrates an example computer system useful for implementing portions of the present disclosure.

[0025] FIG. 15 shows a histogram of expressed high loop gain (HLG) segment lengths across 100 patients with varying levels of self-similarity (SS), including their respective medians. Patient bins were based on the percentage of sleep with automatically detected high SS breathing oscillations. The overall median was approximately 8 minutes.

[0026] FIGs. 16A-16F show scatterplots of the expectation-maximization (EM) algorithm for estimating stability parameters across physiological ranges in simulated breathing data. The algorithm was assessed using three 8-minute simulated breathingsegments, each featuring variably timed airway obstructions. Simulations encompassed all combinations of controller gain (y) values ranging from 0.1 to 2 and delays between the plant and controller (T) from 10 to 40 seconds, across a total of 40 trials per segment. Three initialization approaches were evaluated: (a) initializing parameters to their true simulation values, (b) using a fixed parameter initialization (y = 0.5, T = 15 s), and (c) applying a “warm start,” with five parallel parameter initializations and selection of the optimal result. The shaded area represents the standard deviation of all estimations at each simulated value.

[0027] FIGs. 17A-17B show root mean square error (RMSE) between the simulated and estimated stability parameters, i.e. controller gain (y) and delay between plant and controller (T), across their physiological ranges.

[0028] FIGs. 18A-18F show six 8-minute examples of apnea patients with varying ventilatory control profiles and their respective estimated stability parameters, a-b: 2 segments of a patient showing Cheyne-Stokes respirations with overt centrally driven events and a high loop gain (LG); c-d: segments of patient with predominantly obstructive apneas with respective low and high LG; e-f: segments with mostly hypopneas with an estimated corresponding low and high LG. Estimated parameters for each segment are displayed at the bottom of each subfigure. y=gain, r=delay, Vmax=maximum ventilation rate, L= CO2 production rate, a=disturbance scaling, RMSE=root mean square error between the modelled and the observed ventilation.

[0029] FIG. 19 shows full-night example of patient showing periodic breathing, including estimations for the variable LG parameter for all 8-minute segments of consecutive rapid eye movement or non-rapid eye movement sleep. Note that the respiratory events and high SS breathing oscillations are automatically labelled by our SS detector.

[0030] FIGs. 20A-20C show boxplots of the distribution of estimated loop gain (LG), controller gain (y), and delay (T) parameters across the different ventilatory control profile groups. Compared to the ‘REM OSA’ group all other groups showed a statistically significant increase in median LG / gain. The median T was highest for the heart failure patients. OSA=obstructive sleep apnea, REM=rapid eye movement, NREM=non-REM, CAI=central apnea index, SS=self-similarity.

[0031] FIGs. 21 A-21B show swimmer plots and bar graphs left and right respectively, of 20 individuals from the MGH ventilatory control profiles visualizing the proportion ofelevated loop gain (LG) estimations across overnight recordings. OSA=obstructive sleep apnea, REM=rapid eye movement, NREM=non-REM, CAI=central apnea index, SS=self-similarity.

[0032] FIG. 22 shows scatter plot showing the relationship between loop gain (LG) and self-similarity (SS), including the 5th-95thpercentile range of all data points, from 100 patients. A 2ndorder polynomial was fit to the data together with its 95% confidence interval (CI).

[0033] FIG. 23 shows boxplots of the distribution of estimated loop gain (LG), during rapid eye movement (REM) and non-REM sleep segments from 100 patients with ranging levels of self-similarity (SS). The median LG across all NREM segments was notably higher compared to REM.

[0034] FIGs. 24A-24C show normalized histograms of the estimated LG distribution for 8 patients with each four recordings measured at increasing altitudes, i.e., sea level, 5000 ft, 8000 ft, 13000 ft. The bottom left spaghetti graph shows the trend of the 95thpercentile estimated LG, as denoted by the triangles. The bottom right subfigure shows the automatically detected percentage of sleep with high SS breathing oscillations.

[0035] FIG. 25 shows observed segment of breathing data from respiratory inductance plethysmography and its derived (black) and modelled (blue) components of ventilatory control.

[0036] FIG. 26 illustrates flowchart depicting a method for assessing sleep disorders of a patient, according to embodiments of the present disclosure.

[0037] The features and advantages of the present disclosure will become more apparent from the detailed description set forth below when taken in conjunction with the drawings, in which like reference characters identify corresponding elements throughout. In the drawings, like reference numbers generally indicate identical, functionally similar, and / or structurally similar elements. The drawing in which an element first appears is indicated by the leftmost digit(s) in the corresponding reference number.DETAILED DESCRIPTION OF THE INVENTION

[0038] This specification discloses one or more embodiments that incorporate the features of this invention. The disclosed embodiment! s) merely exemplify the presentdisclosure. The scope of the present disclosure is not limited to the disclosed embodiment(s).

[0039] The embodiment s) described, and references in the specification to "one embodiment", "an embodiment", "an example embodiment", etc., indicate that the embodiment(s) described may include a particular feature, structure, or characteristic, but every embodiment may not necessarily include the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is understood that it is within the knowledge of one skilled in the art to effect such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described.

[0040] Embodiments of the present disclosure may be implemented in hardware, firmware, software, or any combination thereof. Embodiments of the present disclosure may also be implemented as instructions stored on a machine-readable medium, which may be read and executed by one or more processors. A machine-readable medium may include any mechanism for storing or transmitting information in a form readable by a machine (e.g., a computing device). For example, a machine-readable medium may include read only memory (ROM); random access memory (RAM); magnetic disk storage media; optical storage media; flash memory devices; electrical, optical, acoustical or other forms of propagated signals (e.g., carrier waves, infrared signals, digital signals, etc.), and others. Further, firmware, software, routines, instructions may be described herein as performing certain actions. However, it should be appreciated that such descriptions are merely for convenience and that such actions in fact result from computing devices, processors, controllers, or other devices executing the firmware, software, routines, instructions, etc.

[0041] Before describing such embodiments in more detail, however, it is instructive to present an example environment in which embodiments of the present disclosure may be implemented.

[0042] Sleep is a complex state characterized by cycling stages (e.g., rapid eye movement (REM) sleep and non-REM sleep) and a sequence of progressive and regressive depths (e.g., stages I to IV in non-REM sleep). There is also a stability dimension that has been recognized, but not generally applied in clinical practice, based on the concept of cyclicalternating pattern (CAP) in non-REM sleep. CAP -type non-REM sleep is unstable, whereas, non-CAP non-REM sleep is a restful, non-aroused state with a stabilizing influence.

[0043] The prevalence of sleep-disordered breathing (SDB) among adults has increased substantially in recent years, in tandem with the prevalence of obesity. These sleep disorders are significant risk factors for cardiometabolic and neurodegenerative diseases, impaired performance, and decreased quality of life. One of most widely used method to help sleep disorders is continuous positive airway pressure (CPAP). However, tolerance and efficacy of CPAP are often poor. Among several reasons for CPAP intolerance, one area that stands to benefit from the computational analysis is the precision phenotyping of sleep-breathing patterns.

[0044] The embodiments are directed to developing a computational approach to detect HLG based on self-similarity in respiratory oscillations during sleep solely using breathing patterns, as measured via respiratory inductance plethysmography (RIP).

[0045] The embodiment are further directed to detecting apneas and hypopneas as periods with reduced breathing effort, manifested in the RIP signal as low signal amplitude and calculating self-similarity in breathing patterns between consecutive periods of apnea or hypopnea.

[0046] As described in greater detail below, the present disclosure provides techniques and / or methodologies for assessing signals related to breathing during sleep and determining HLG obstructive sleep apnea. Accordingly, the present disclosure provides an improved away of detecting patients with HLG obstructive sleep apnea to adequately assisting.

[0047] FIG. 1 illustrates an example diagram of a patient 102 having a sensor 104 placed on patient 102, according to embodiments of the present disclosure. The sensor may be placed on a patient body to measure breathing activity or breathing motion of patient 102 over a period of time while sleeping. Any known device can be used for measuring breathing motion. For instance, a band having two sinusoid wire coils may wrap around the patient near patient’s chest, and the band have a small current to detect expansion and contraction of the chest to measure the change in lung volume. Other examples include a blood pressure sensor, a flowmeter, an anemometer, a fiber optic sensor, a microphone, a thermistor, a thermocouple, a pyroelectric sensor, a capacitive sensor, a resistive sensor, ananocrystal and nanoparticles sensor, infrared, inductive, transthoracic, inertial, an electrocardiogram sensor, a photoplethysmography sensor, etc. It should be understood that the exact placement of the leads is not intended to be limiting.

[0048] A system based on RIP tracings may be for automated phenotyping during routine polysomnography (PSG) recordings and complementary to any manual approach to phenotyping for detecting apneas and hypopneas as periods with reduced breathing effort, manifested in the RIP signal as absent or low signal amplitude.

[0049] The present disclosure involved comparison of detect expressed HLG independent of manual scoring of obstructive against central respiratory events.DATA PREPARATION

[0050] Split-night recordings may be stratified into three groups based on their residual apnea-hypopnea index (AHI) and central apnea index (CAI) during the titration phase of each patient. The time up to CPAP administration may be considered the diagnostic phase, whereas the latter part of the recording may be considered the titration phase. At least one hour of sleep may be required in each study phase. CPAP treatment success or failure can use several reasonable thresholds, for example, CPAP may be considered to be successful if it results in a combined AHI<10 and CAI<5, while a residual AHI>30 orCAI>10 may be clearly considered therapy failure. Other thresholds may also be usefully computed, with an impact on specificity and sensitivity. Any recording outside these two groups may be deemed to be undefined CPAP effectiveness. The AHI and CAI metrics may be computed as follows:where OA, CA, MA, and HY refers to obstructive apneas, central apneas, mixed apneas and hypopnea, respectively.

[0051] While the American Academy of Sleep Medicine (AASM) previously considered a nasal pressure (or PAP flow) signal excursion drop of >30% together with a > 4% oxygen desaturation from pre-event baseline a hypopnea, now similar excursion drops with corresponding >3% oxygen desaturation and / or arousal are recommended to be scored as hypopnea. The latter type of hypopnea events may be added to dataset by using an automated respiratory event scoring algorithm (40).MEASURING SELF- SIMILARITY ACROSS BREATHING OSCILLATIONS

[0052] To measure the level of SS in effort signals, present waxing and waning breathing oscillations may be identified. By connecting consecutive inspiratory and expiratory amplitude peaks the ventilatory envelope may be computed, allowing to identify local reductions in breathing effort. During sleep, amplitude reductions >30% with a duration of at least 10 seconds may be associated with hypopneic events while apneic events typically require 90% or greater signal amplitude reduction. FIGS. 2A-2F show how both a trace with regular breathing and a trace with waxing and waning breathing oscillations may be evaluated using their respective envelopes. In particular, two example effort traces, the upper one with regular breathing, and lower one with self-similar breathing oscillations, in which inspiratory and expiratory peak envelopes were used to find amplitude reductions of at least 30%, with a minimum duration of 10 seconds.

[0053] Areas between effort reductions may be tagged as breathing oscillations if the distance between the consecutive effort reductions may be less than two minutes, ensuring waxing-waning breathing oscillations with a short (<30 seconds) or long (>60 seconds) cycle time can be captured (21). To differentiate the self-similar waxing and waning phenotype from reductions in effort related to classic OSA, envelope segments may be included when meeting both of the following requirements:1. the positive and negative (180° rotated) envelope have a convolution score >0.7; and2. the location of maximum distance between the positive and negative envelope (tmax) is <15%, relative to the respective midpoint between the two effort reductions, as below: 0.15,where ti and t2 represent the distance to tmax from the leading and trailing effort reductions, respectively (see FIG. 3).PREDICTING CPAP EFFECTIVENESS USING SELF- SIMILARITY

[0054] For each patient, the diagnostic phase of their sleep recording may be cut into five-minute segments. In each segment in which the patient is asleep for at least one minute, the average SS score across all tagged breathing oscillations may be calculated. Segments without breathing oscillations may be set to a score of zero. To summarize thelevel of SS across the recording, histograms may be computed using 10 bins. All histograms may be normalized, so that the total of all bins adds up to 100. In each training set, the average histogram for both the success group and the failure group may be calculated by adding all respective histograms together and dividing them by the number of patients within their respective groups.

[0055] Each patient histogram from the test set may be compared to both the average histogram from the success and failure group, resulting in two error scores:1. comparing a patient histogram to the average histogram of all training patients with little to no residual respiratory events; and2. comparing a patient histogram to the average histogram of all training patients that showed significant residual respiratory events.

[0056] The mean absolute error (MAE) may be computed using the following equation:where n is the number of histogram bars (10), and, hteand htr represent the histogram of the test patient and the average histogram of the training patients, respectively. The difference between the two error scores may be used as the input for the logistic regression model.METHODSDataset description and preparation

[0057] Dataset was selected from an archive containing clinical polysomnography (PSGs) recorded at the Massachusetts General Hospital (MGH) sleep laboratory between 2008 and 2022. The MGH Institutional Review Board approved the retrospective analysis of clinically acquired PSG data. All sleep studies included abdominal respiratory inductance plethysmography (RIP) signals with a sampling frequency of 200 Hz or 512 Hz. A notch filter of 60 Hz (power line) and a lowpass filter of 10 Hz were applied to reduce non-physiological noise. Subsequently, recordings were resampled to 10 Hz and normalized using the mean and standard deviation of the 5th-95th percentile clipped signal.

[0058] The MGH sleep center is an American Academy of Sleep Medicine (AASM) accredited sleep center that scores obstructive, central, and mixed apneas, hypopneas (using the 4% desaturation rule), and respiratory effort-related arousals (39). To conformevent scoring with current AASM convention, hypopneas meeting the 3% / arousal criteria were added using an automated scoring algorithm (40).Estimation of titration success / failure

[0059] To create clear success and failure categories, recordings with an apnea-hypopnea index (AHI) >10 before CPAP were categorized based on their AHI and central apnea index (CAI) during the titration phase of CPAP into a therapy success group (AHI <10 and CAI <5) and a therapy failure group (AHI >30 or CAI >10 despite CPAP). Recordings outside these two groups were excluded. To visualize the relationship between the number of respiratory events and CPAP effectiveness, histograms were created showing the fraction of patients that fail therapy across the range of CAI, AHI, and SS.Predicting CPAP effectiveness

[0060] Respiratory SS was measured by identifying local fluctuations in breathing effort. Ventilatory envelopes were computed by connecting inspiratory and expiratory amplitude peaks. Breathing oscillations with the self-similar waxing and waning phenotype were differentiated from obstructive events by evaluating envelope segments based on two requirements: high horizontal symmetry between positive and negative breathing envelope, and a point of peak excursion close to the respective midpoint (FIGS. 2A-2F and 3). Scores ranging from 0-1 were assigned to tagged breathing oscillations, reflecting a scale from low to high SS. See the supplemental material for an in-depth technical description. Below, the mathematical representation of the two morphology tests are denoted. FIGS. 2A-2F show visual representations of the two tests.

[0061] Test 1 : The normalized positive (pnorm) and 180° rotated negative (nnOrm) envelope show a convolution score exceeding a threshold of 0.5., for example, the segment length and the positive / negative. , . .. . . envelope (x) are normalized by

[0062] Test 2: The location of maximum distance between the positive and negative envelope (tmax) is <15%, relative to the respective midpoint between the two effort reductions, for example, , where ti and t2 represent the distance to tmax fromthe leading and trailing effort reductions, respectively. For instance, FIG. 3 illustrates anexample of how envelopes of breathing oscillations were assessed by (1) computing a convolution between the positive and the rotated negative envelope, and (2), assessing the location of the maximum distance between the positive and negative envelope (tmax).

[0063] Referring to FIG. 4, which illustrates an example of how self-similarity in breathing is computed using convolution scores across neighboring envelopes of oscillations, 3 -minute breathing segments that includes three potential breathing oscillations, meaning each oscillation exceeds both morphology tests, was assessed for SS. A breathing oscillation was marked as expressed high loop gain when the maximum envelope convolution score among its 2 two neighboring oscillations (SSy,x,z) exceeds a user-defined threshold ranging from 0-1 (default=0.8), for example,max

[0064] The SS scores during the diagnostic phase were summarized across five-minute segments. Histograms were generated from these scores, and a logistic regression (LR) model was trained and evaluated using 5-fold cross-validation. For each test patient two mean absolute histogram distances were computed, first with reference to the average histogram of all training patients with a successful therapy outcome, and second, with reference to the failure group; for details see the supplemental material. The difference between the histogram distances, summarized into a single error score, was used as input for the LR model. In addition, LR models were trained using the CAI and the hypoxic burden as individual inputs, and in combination with SS.

[0065] Models were evaluated for accuracy and calibration on the entire cohort and on a subpopulation of patients with a CAI <1. Receiver operating characteristic (ROC), precision -recall (PR) (including 95% confidence interval using bootstrapping, N= 10,000), and calibration curves were computed. Heatmaps were used to study the relationship between the level of SS and the hypoxic burden.RESULTSPatient characteristics and algorithm

[0066] Data from 2145 unique split-night recordings were included and are summarized in Table 1.Table 1 : Cohort demographics (N=2145)

[0067] Examples comparing the new algorithm with the previously published method are shown in FIGS. 5A-5F which show example effort traces with technician scored obstructive events and central events. The asterisk symbol (*) and circle shape (°) indicate breathing oscillations marked as high loop gain by the old and improved self-similarityalgorithm, respectively. In FIGS. 6A-6D, histograms and example breathing segments from four patients with varying levels of SS are shown. For instance, in FIGS. 6A-6D, respiratory effort example traces from 4 patients ranging from low to high SS, from left to right. The associated histograms show how self-similarity (SS) across all 5-minute segments was summarized for these patients. Tagged high loop gain breathing oscillations are marked with asterisk symbol (*). These show that as the level of SS increases, the number of tagged HLG breathing oscillations increases, and the distribution becomes increasingly skewed rightward. Referring to FIG. 7, which illustrates a pie chart of the proportion of tagged high loop gain breathing oscillations per sleep stage, over 95% of all the SS oscillations were found in NREM sleep.Self-similarity and CPAP

[0068] The proportion of patients who continued to experience apnea and hypopnea during CPAP was positively correlated with higher baseline values of CAI, AHI, and SS (FIGS. 8A-8C). Histograms showing the fraction of patients who fail CPAP therapy across the range of the apnea-hypopnea index (AHI), the central apnea index (CAI) and self-similarity (SS), based on the 3% hypopnea rule. Sex proportions are displayed in red / blue. The number above each bar indicates the sum of patients contributing to each bin High SS was more prevalent in men; most patients who acutely failed CPAP were also men, with the exception of individuals with an AHI >125, whereas the CPAP success groups show a more even distribution. FIGS. 9A and 9B show two overnight recordings from patients exhibiting mild and severe SS, respectively. Each recording was cut into 10 consecutive RIP breathing segments from left to right and top to bottom.

[0069] The predictability of CPAP failure using SS, CAI and hypoxic burden as features to predict CPAP failure is shown in the ROC and PR curves in FIGS. 10A-10F. FIGS. 10A-10F illustrate Receiver Operating Characteristic (ROC), Precision-Recall (PR), and calibration curves of the central apnea index (CAI), hypoxic burden (Burden), and selfsimilarity (SS) as individual predictive features for CPAP failure. Below the performance of each feature combination is shown. CPAP treatment effectiveness is computed using the 3% hypopnea rule. Area under the curve (AUC) values and 95% confidence intervals (CI) for the ROC and PR curves are added in their respective legends.

[0070] The area under the curve (AUC) values obtained by using SS, CAI and hypoxic burden as individual features were 0.75 [0.73, 0.78], 0.73 [0.70, 0.75], 0.65 [0.63, 0.68],and 0.76 [0.73, 0.80], 0.77 [0.74, 0.79], 0.69 [0.65, 0.72] for the ROC and PR curves respectively. The new SS algorithm showed significantly higher AUC values compared to the original algorithm. AUC values further increased when combining predictive features, up to an ROC AUC of 0.82 [0.80, 0.84] and a PR AUC of 0.83 [0.81, 0.86] when combining SS with the CAI and hypoxic burden as three input features for a LR model. All LR models showed excellent risk calibration (calibration curves are close to the diagonal). It is notable that for the individual features, no calibration points below a probability of 0.3 were observed. When combining features, the range of prediction increases and calibration points between 0.2 and 0.9 were observed. .

[0071] Using the models trained on the entire cohort, patients were examined with a CAI <1 (N=455). This showed a clear decrease in performance when using CAI alone as a predictive feature. FIGS. 11 A-l IF shows that SS did provide ROC and PR AUC values of 0.73 and 0.57, respectively, outperforming the CAI and hypoxic burden. For instance, FIGS. 11 A-l IF illustrate Receiver Operating Characteristic (ROC), Precision-Recall (PR), and calibration curves of the central apnea index (CAI), the hypoxic burden (Burden), and self-similarity (SS) for all patients with a pre-treatment CAI <1 (N=455). The upper three graphs show the individual negative predive features for CPAP therapy, whereas the performance of each feature combination is shown below. Area under the curve (AUC) values and 95% confidence intervals (CI) for the ROC and PR curves are added in their respective legends. It is observed that patients with treatment-emergent central apnea and high residual obstructive events, while being labeled with a CAI of zero before titration, are better identified using SS.

[0072] FIGS. 12A and 12B illustrate Heatmaps showing the relationship between the level of self-similarity and the hypoxic burden within the CPAP success and CPAP failure groups, left and right respectively. A clear cluster of patients with low SS and a low to moderate hypoxic burden is apparent within the CPAP success group, whereas for patients from the CPAP failure group clusters with low and high SS spanning the entire range of the hypoxic burden were observed.DISCUSSION

[0073] This work demonstrates that the risk of acute CPAP failure is predictable from the morphology of breathing signals. The described algorithm aims to detect expressed HLG - using the morphological patterns shared in common across periodic breathing, centralsleep apnea and NREM-dominant HLG driven apnea (41). While presently available algorithm showed reduced accuracy in scenarios with subtle asymmetry across consecutive breathing oscillations, the morphology assessment in the present work improves sensitivity such that HLG-driven events are now better detected.Central apnea as predictor of CPAP effectiveness

[0074] The central apnea index does not accurately predict residual respiratory events during CPAP for all patients. FIG. 7 shows that for an increasing CAI, patients have a decreasing probability of sufficiently benefitting from CPAP, yet a significant subgroup of patients with few central apneas continue to experience respiratory events during titration. The results show that a patient’s risk of persistent apneas and hypopneas cannot be fully captured by scoring central apneas alone, and accurate scoring of central hypopneas have remained elusive. Drive-dependent respiratory events dominate in obstructive sleep apnea (42), so it should be no surprise that central hypopneas are also associated with flow-limitation (43). Thus, manual / visual scoring of individual respiratory events cannot accurately determine the pathophysiological driver of hypopneas. The rigidity of the AASM scoring rules for central hypopneas exclude any obstructive feature, while it is well documented that central apneas can have a closed airway and even at high altitude flow-limitation respiratory events are commonly seen (44, 45).Morphological CPAP success / failure prediction

[0075] Improved prediction of CPAP effectiveness is obtained in this present approach by identification of expressed HLG in respiratory effort signals. The level of morphological SS across breathing oscillations as an individual predictor achieves similar accuracy compared to CAI when this index is high, with ROC and PR AUC values of 0.75 and 0.76 respectively, including accurate calibration, see FIGS. 8A-8C.

[0076] The SS analysis is however most useful when central apneas do not dominate in a given patient, but central hypopneas, events with some flow-limitation, and physiologically mixed events are frequent. The struggle to differentiate central from obstructive hypopneas may be mitigated and effectively bypassed by this method. Events detected by SS analysis could be called “HLG-events” or something along those lines that reflect driver pathophysiology, to move away from the scoring tension betweenobstructive and central hypopneas. While a high hypoxic burden may indicate that a patient's sleep apnea is more severe, SS also outperformed hypoxic burden for prediction of residual events, as the latter showed AUC values ranging between 0.62-0.72. When combining SS with CAI and hypoxic burden, higher predictive performance was obtained with AUC values of approximately 0.82-0.83, suggesting that the three metrics are complementary.Hypoxic burden and CPAP failure

[0077] Intermittent hypoxia increases loop gain through carotid body sensitization (46), and CPAP reduces loop gain in patients with hypoxic sleep apnea (9). A low SS and low to moderate hypoxic burden may predict CPAP being effective, yet many patients with this phenotype still experienced unsuccessful treatment, as observed in FIGS. 12A and 12B . Moreover, patients with a low or high hypoxic burden were likely to continue to experience respiratory events during CPAP if they display high SS in their breathing. These findings suggest that patients with HLG, regardless of oxygen desaturation, have a large positive likelihood of CPAP being acutely ineffective. In fact, desaturation is often mild in idiopathic central sleep apnea and heart failure associated periodic breathing due to the vigorous arousals and hyperventilation inherent in these disorders (47, 48). Thus, events driven by HLG are related to poor therapy outcomes, even though such events do not necessarily cause significant desaturation.Comparison to the published literature

[0078] Loop gain estimates are not yet part of standard clinical practice due to the lack of easily applied methodology. Various publications evaluate the ventilatory response and sensitivity to hypoxia, by administering a gas mixture containing adjusted O2 / CO2 levels (5-9). These methods require monitoring of the gas concentrations in inhaled and exhaled air while typically only flow, pressure, and effort breathing signals are obtained sleep during conventional sleep studies. Controlled mechanical ventilation can also be used to assess LG. The response of the patient’s respiratory system to changes in tidal volume and respiratory rate provided by the ventilator can be measured and used to calculate LG (10, 11, 49).

[0079] Non-invasive LG assessment is possible by applying breath-holding maneuvers. However, this technique requires active participation from awake individuals whenbreathing control and breathing stability can be significantly different compared to sleep (12-14). The literature also describes techniques that use the response to spontaneous respiratory events to measure LG by leveraging mathematical parameter estimation of the underlying breathing mechanics (15-17). While this method can be used with standard PSG signals, the need for manual signal preparation and event labelling currently prevents this technique from clinical adoption. However, at least one vendor of sleep testing equipment now integrates automatic endotype assessment from effort signals. Cardiopulmonary coupling can detect HLG by identifying the presence of narrow-band coupling, another expression of self-similarity (18, 19).

[0080] However, the technique requires 15-17 continuous minutes of self-similar cardiopulmonary coupled oscillations for detection, thus brief bursts of periodic breathing are not detected. Table 2 shows an overview of existing LG measurement techniques including their main advantages and disadvantages.Table 2: Overview of previously published high loop gain (HLG) measurement techniques including their main advantages and disadvantages

[0081] Ultimately, multiple complementary methods applied to individual patient data may be best. The described method can be easily applied to signals from laboratory or home polysomnograms; the goal is to have this available for common use as an aid to clinical risk stratification.Limitations

[0082] The long-term effects of detected SS on CPAP outcomes was not estimated in this study and is a next step in the research program. However, there is published data showing that central sleep apnea at baseline, reflecting high loop gain, predicts residual ventilatory instability of positive pressure therapy. Split-night polysomnograms were chosen to enrich for disease severity; moreover, the whole range of possible combinations of AHI and CAHI were not evaluated. Night-to-night variability of endotypes and stability over time requires further study (50). The current approach only identifies expressed HLG, making mildly elevated LG in apnea patients undetectable. Loop gain was not directly measured but emergent central apneas and SS events were used as surrogate markers of HLG.

[0083] The analysis is currently solely based on effort signals via RIP, whereas other signals, e.g., oximetry morphology, the ballistocardiogram, the electroencephalogram, the electromyogram, beat to beat blood pressure, the arteriolar photoplethysmographic signal, and airflow may convey additional relevant information. The results from this study show, however, that using effort signals only can differentiate patients who will continue to experience significant respiratory events during CPAP from patients with effective therapy with sufficient clinical accuracy. A major advantage of using an effort belt is that it allows for non-invasive application such as home monitoring and monitoring in the intensive care setting.

[0084] While the described embodiment is in relation to positive airway pressure therapy, it can be readily applied to risk stratify outcomes from any form of obstructive sleep apnea treatment, including oral appliances, various surgical procedures to improve upper airway anatomy, hypoglossal or ansa cervicalis stimulation, myofunctional therapy, neuromuscular electrical tongue stimulation such as eXciteOSA, and even pharmacotherapy for sleep apnea.

[0085] While the described embodiment is in relation to sleep, periodic breathing (and thus self-similarity of ventilation) can occur during wake at extreme altitude or severe heart failure, and during exercise (or exercise testing), where it is called oscillatory exercise ventilation (OEV), with prognostic (adverse) implications in cardiopulmonary disease (51-58). The mechanisms include an expression of high loop gain, but during thewake state. The SS detection principle can be applied to estimating / quantifying / tracking OEV also.CONCLUSION

[0086] A clinically applicable method to detect expressed HLG independent of manual scoring of obstructive vs. central respiratory events is described. This estimate accurately predicts acute failure of CPAP, which itself is associated with high residual apnea and CPAP non-adherence. Self-similarity may help risk-stratify, and identify individuals who could benefit from adjunctive therapy targeting HLG, particularly when affected by centrally driven events that do not meet the criteria for CSA.

[0087] FIG. 13 illustrates flowchart depicting a method 1300 for assessing sleep disorders of a patient, according to embodiments of the present disclosure. It is to be appreciated that method 1300 may not include all operations shown or perform the operations in the order shown.

[0088] Method 1300 begins at step 1302 where signals form a patient is received while the patient is asleep. The signals may be representative of sleep motion of the patient, e.g., by measuring lung expansion and contraction over a period of time as described above with reference to FIG. 1.

[0089] At step 1304, a complex within the received signals may be identified. For example, an envelope of the received signals over the period of time may be ascertained, as described above with reference to FIG. 2D.

[0090] At step 1306, a ratio of a maximum envelope height to a minimum envelope height within the complex may be determined. For example, referring FIG. 2E, maximum and minimum envelope heights may be identified in each complex.

[0091] At step 1308, a duration between a maximum envelope height of the complex and a subsequent maximum envelope of a subsequent complex may be determined. For example, referring to FIG. 2F, the duration between two peak signals may be determined.

[0092] At step 1310, a convolution of a positive envelope and a rotated negative envelope of the complex may be computed. The rotated negative envelope may be a rotation of the negative envelope about a time axis as shown. For example, referring to FIG. 3, in step (1), the maximum height and the minimum height, which may be convoluted and compared with the maximum height for similarity analysis.

[0093] At step 1312, a time location of a maximum distance between the positive envelope and the negative envelope may be assessed. Referring to FIG. 3, in step (2), the time point of the peak signal may be determined, and a duration from one minimum peak to next maximum peak and a duration from this maximum peak to the next minimum peak may be determined.

[0094] At step 1314, a self-similarity of the received signals may be determined based on steps 1302 through 1310. For example, the determined values from steps 1302 through 1310 may be computed to determine a self-similarity.

[0095] Data obtained for a period of time when a patient’s sleep were assessed for each complex as there are a number of complexes, for example, as shown in FIGS. 5A-5F, and it is possible to predict the probability of failure (high residual event counts) of continuous positive airway pressure (CPAP) therapy with the above-described method.ADDITIONAL APPROACHES

[0096] Sleep apnea is a prevalent disorder affecting millions globally, with profound implications for cardiovascular health, cognitive function, and overall quality of life [1,2,57], Accurate identification of apnea phenotypes, such as obstructive sleep apnea (OSA) and central sleep apnea (CSA) is critical for effective treatment [58,59], The need for precise assessment extends beyond polysomnography (PSG) and home sleep apnea testing, including ventilator weaning, wearable technology monitoring, and cardiopulmonary exercise testing, particularly in cases of exercise oscillatory ventilation (EOV).

[0097] Ventilatory control, which is influenced by factors like chemoreflexes, lung mechanics, and arousal thresholds, determines how the body responds to changes in oxygen and carbon dioxide levels, and plays a crucial role in the severity of apnea [4,9,60], A key measure of this control is ventilatory loop gain (LG), which indicates the sensitivity of the respiratory system to these changes. A high loop gain (HLG) signifies a more reactive system, potentially leading to breathing instability. In OSA, HLG can exacerbate respiratory instability and complicate interventions like ventilator weaning [3], In CSA and Hunter-Cheyne-Stokes respiration (HCSR), increased chemosensitivity (higher controller gain) combined with delayed responses can hinder the body's ability to stabilize breathing, thereby prolonging episodes of unstable breathing and apnea

[0061] ,

[0098] Recognizing the unique role of LG in apnea subtypes is essential for tailored interventions. For example, OSA patients with average LG should respond to continuous positive airway pressure (CPAP), whereas HLG patients may require additional O2, CO2 therapy, acetazolamide, zonisamide or sulthiame [32-35,62], Beyond OSA, LG assessments can inform strategies for ventilator management and optimize treatment in patients based on their breathing stability.

[0099] Despite the availability of several methods for estimating LG in apnea patients, including the use of hypoxic-hypercapnic gas [5-9], breath-holding techniques [12-14], mechanical ventilation [10,11,49], cardiopulmonary coupling [18,19], and mathematical modelling of breathing dynamics [15-17], practical challenges such as manual data annotation and the need for specialized equipment have limited their clinical use. To overcome these barriers, we recently developed an automated algorithm that detects HLG by detecting self-similarity (SS) in the envelope morphology of abdominal respiratory inductance plethysmography (RIP) signals, eliminating the need for manual processing [20,63], While this method is fully automatic, it is indirect, focusing on detecting expressed HLG rather than the entire spectrum of LG intensities or the physiological interplay of its underlying parameters.

[0100] In this study, we present a model-based framework for automated LG assessment based on RIP signals. Our approach uses modified Mackey-Glass equations to model ventilatory control as a closed-loop system, enabling estimation of key parameters including gain and delay. By providing a practical method to quantify ventilatory control stability, this work aims to support more targeted treatment selection in sleep apnea and related respiratory disorders.METHODSDataset

[0101] The primary data used in this work were selected from an archive containing clinical PSGs from the Massachusetts General Hospital (MGH) sleep laboratory between 2008 and 2022. The MGH and the Beth Israel Deaconess Medical Centre (BIDMC) Institutional Review Boards approved the retrospective analysis of clinically acquired PSG data with waiver of informed consent (protocol #s: MGH: 2013P001024, BIDMC: 2016P000058). The MGH sleep centre is an American Academy of Sleep Medicine(AASM) accredited sleep centre that scores the various apnea endotypes and hypopnea’s using the 4% desaturation rule. Hypopneas meeting the 3% desaturation or arousal criterion were automatically added using a previously developed respiratory event detector

[0040] , A secondary dataset with heart failure (HF) patients was used to study the known association between HLG and this specific population [64,65], Thirdly, PSG studies from BIDMC recorded four incremental altitudes were included to show the relationship between LG and high-altitude breathing.Patient selection based on ventilatory control profile

[0102] From all MGH recordings with an apnea hypopnea index (AHI) >15 and a minimum of 4 hours of sleep, four patient subgroups were created based on their ventilatory control profile: 1) a “REM OSA” group including patients in which >10% of their sleep was rapid eye movement (REM) combined with a REM AHI being at least three times higher than the AHI during non-rapid eye movement (NREM); 2) a “NREM OSA” group including patients in which >10% of their sleep was REM combined with a NREM AHI being at least double their AHI during REM; 3) a “SS OSA” group with a central apnea index (CAI) <5 and in which >25% of their sleep was detected as having high SS; 4) a “high CAI” group including patients with a CAI >5, and a CAI > OAI. Finally, a “SS range” group of apnea patients ranging from low too high levels of SS were selected. 100 recordings were sampled for each subgroup. For the secondary dataset, all patients with an AHI >10 and an ejection fraction <50 were included, collectively denoted as the “Heart Failure” group.Preprocessing

[0103] Abdominal RIP signals were extracted with a sampling frequency of 200Hz or 512Hz. After applying a notch filter of 60Hz (to remove power line noise) and a lowpass filter of 10Hz to reduce non-physiological noise, all recordings were resampled to 10Hz. Next, recordings were normalized using the mean and standard deviation of the 5th-95thpercentile clipped signal. Ventilatory envelopes were computed by connecting inspiratory and expiratory amplitude peaks, from which reductions of 30% and 90% for a duration >10 seconds were associated with hypopnea and apnea, respectively. Surrogate ventilation traces were computed by calculating the distance between the positive andnegative envelope. Sharp increases in ventilation were saved as potential arousal locations, generally located at the end of respiratory events.Segmentation

[0104] Characterizing ventilatory control overnight is difficult as its dynamics typically fluctuate

[0066] , To assess the fragmented nature of expressed HLG in patients, we used our published SS algorithm in combination with a change-point detector

[0067] to approximate the most common duration of expressed HLG in breathing. Using the patients in the “SS range” group, we created a histogram of the lengths of all high SS chunks and used its median (8 minutes) for further analysis.Modelling and estimation of ventilatory control

[0105] For each 8-minute segment of breathing data we modelled the local dynamics of ventilatory control.

[0106] The Mackey-Glass equations serve as the cornerstone of our methodology:which states that the rate of change of the arterial CO2 concentration, x, is the CO2 production rate, L, minus the ventilation rate a brief time T in the past, vd(t — T), times the current CO2 concentration. Here, T is the delay between plant (the lungs) and the controller (the brainstem). The ventilation rate vd(t) in turn is modeled as a sigmoid curve with two parameters: Vmax, the maximum ventilation rate, and y the controller gain (steepness of the sigmoid curve), such that ventilation increases with higher levels of CO2. Building upon Equation 1, we introduce a parameter for breathing disturbance (U), corresponding to the reduction in ventilation during respiratory events (apneas or hypopneas), and a parameter for arousals (yar), which cause transient increases in ventilation. With these modifications, we rewrite the original Mackey-Glass equations as:

[0107] In this ventilatory control model we assume a constant CO2 production rate (L = 0.05) and we derived U and varfrom the surrogate ventilation traces. All other parameters, i.e , and the height of each individual arousal, are treated ashidden parameters to be estimated. To estimate these hidden parameters, we used the expectation-maximation (EM) algorithm

[0039] , Initially, in the E step, this algorithmestimates the hidden parameters by computing their probability distribution based on the observed ventilation data from patients. Subsequently, in the M step, it refines the model's parameters by maximizing the expected log-likelihood computed in the E step. By iteratively switching between these steps, the EM algorithm optimizes the model's parameters, by reducing the root mean square error (RMSE) between the modelled v0and observed ventilation derived from the RIP signals.

[0108] Given the estimated parameters, we compute LG by analysing the system's response to disturbances and subsequent recovery. Specifically, LG is computed as the ratio of the change in ventilation immediately following the end of an apnea or hypopnea (the “disturbance”), (AVR), to the change in ventilation induced by the disturbance (AVD). Here, AVR is the absolute difference between the steady-state ventilation AVss and the ventilation after removing the disturbance, while AVD represents the absolute difference between AVss and the ventilation during the disturbance. This ratio, LG = AVR / AVD, quantifies the system’s feedback sensitivity, determining whether the system maintains stability or exhibits oscillatory behaviour. For more mathematical details see the supplemental material.Simulation and validation of the LG estimator

[0109] Simulations assessed algorithm accuracy by comparing estimated parameters with true values, varying y (0.1-2.0) and T (5-50 seconds) to evaluate robustness across physiological ranges. Error graphs summarized estimation performance.

[0110] Boxplots displayed parameter distributions for REM OSA, NREM OSA, SS OSA, and high CAI groups, with T-tests comparing REM OSA to other subgroups for LG and y. A scatterplot with a 2ndorder polynomial fit (95% confidence interval with 10,000 bootstraps) of the “SS range” group, including its r2value and 5th-95thpercentile range. Boxplots were created showing the distribution of the estimated LG within NREM vs. REM within the “SS range” group. For the altitude data, histograms of the estimated LG distribution were created. Unpredictable LG segments (fewer than 5 apneas / hypopneas in 8-minute windows) were excluded. The 95thpercentile LG value per patient highlighted LG trends across increasing altitudes, minimizing regular breathing effects and outliers.RESULTSDataset[OHl] Data from MGH and BIDMC are summarized in Table 3. The "Heart Failure" group included 65 patients. Altitude data from 8 patients with recordings at sea level, 5, 8, and 13 thousand feet were analysed. High SS chunks (expressed HLG) in the “SS range” group, summarized in a histogram (FIG. 15), typically lasted 4-10 minutes, with a median duration of approximately 8 minutes.Table 3: Cohort demographics (mean ± SD).OSA, obstructive sleep apnea. CAI, central apnea index. AHI, apnea hypopnea index. SS, self-similarity.Simulation experiments

[0112] We compared the estimated parameters y and r against known simulation settings. Initializing the EM algorithm with true values resulted in highly accurate estimates (FIGs. 16A-16F, grey line). To evaluate performance in non-simulated clinical scenarios, we assessed two initialization approaches: fixed parameter initialization and a “warm start” (using 5 parallel initializations and selecting the best result). Both approaches demonstrated similar performance across three simulations. Strong positive correlations were observed between median estimated and simulated y values, though higher y values showed increasing underestimation. Median T estimations were nearly perfect across all simulations. Due to 5-fold faster computation, fixed parameter initialization was selected for further research. Contour plots of RMSE for simulated vs. estimated y and T values (FIGs. 17A-17B) show that lower y estimates yield higher accuracy, while RMSE for T remains below 2 across its physiological range.Modelling ventilatory control

[0113] FIGs. 18A-18F depict six 8-minute breathing segments from patients suffering from various apnea phenotypes. From top to bottom, the figures show two segments of a patient exhibiting HCSR with overt centrally driven events and HLG; 2 segments of patient with predominantly obstructive apneas with respective low and high estimated LG; 2 segments with mostly hypopneas with a corresponding low and high estimated LG. Corresponding LG values were found after estimating the underlying hidden dynamics that resulted in the best fit with the patient’s breathing data, i.e., the parameter combination of y, T, a, and Vmax that minimized the RMSE between the overserved and modelled ventilation. Note how the modelled vdclosely resembles the patient’s ventilation trend and varmatches the peak ventilation amplitude following respiratory events. FIG. 19 shows a full-night RIP recording of a patient with HCSR including the varying LG estimations for all 8-minute segments of consecutive NREM or REM sleep (without intermittent wake). Note that the respiratory events and high SS breathing oscillations are automatically labelled by our SS detector.LG across patient ventilatory control profiles

[0114] Boxplots in FIGs. 20A-20C show the distribution of the estimated LG, y, and T values across patient subgroups. Patients from the “REM OSA” group show the lowestmedian LG. Relative to this group, all other groups showed higher median LG and y values with strong significance (p<0.001). Notably, patients from the “high CAI”, “SS OSA” and “Heart Failure” ventilatory control profile groups showed the largest overall LG. A median T of 26 seconds was found for the “Heart Failure” patients, whereas median values closer to 20 seconds were found for the other apnea subgroups. Swimmer plots and bar graphs of the LG estimations from 20 nights of each of the ventilatory control profiles of the MGH dataset are visualized in FIGs. 21 A-21B. Compared to the “REM OSA” group, the proportion of HLG among the other ventilatory control profiles is markedly increased.SS vs LG

[0115] To characterize the relationship between the estimated LG and expressed LG, as measured through self-similarity (SS), FIG. 22 shows a scatterplot including the 5th-95thpercentile range. A 2ndorder polynomial fit to the data has an r2value of 0.75, indicating a strong monotonic relationship. The LG distribution across all NREM segments showed a higher median compared to during REM sleep, shown in FIG. 23.Altitude vs LG

[0116] Estimations from our LG model for 8 patient recordings at varying altitudes are shown in FIGs. 24A-24C. The upper section displays normalized histograms of LG distributions at sea level, 5000 ft, 8000 ft, and 13,000 ft. The bottom left graph shows an increasing trend in the 95thpercentile LG (triangles), while the bottom right graph depicts the percentage of high SS (expressed HLG) breathing oscillations. Both indicate higher mean and median LG at elevated altitudes compared to sea level. Interestingly, for some patients (e.g., 2 and 5), HLG levels were higher at 5000 / 8000 ft than at 13,000 ft.DISCUSSION

[0117] This study presents a novel method for estimating LG and its ventilatory control parameters using RIP signals. By applying modified Mackey-Glass equations to overnight breathing data, the approach offers a practical, equipment-free assessment of ventilatory stability. Our results indicate its potential to differentiate LG patterns in diverse sleep-disordered breathing populations. This capability, using routine clinicalsignals, could guide treatment selection and improve outcomes, especially in cases where ventilatory control instability exacerbates sleep apnea severity.

[0118] LG measurements offer valuable potential for clinical decision-making in sleep apnea treatment. HLG can predict CPAP intolerance, inadequate treatment response, and persistent residual apnea, allowing clinicians to stratify patients by risk and optimize therapy selection [28-30,63], When considering alternative treatments such as oral appliances or hypoglossal nerve stimulation, LG assessment can identify patients who may require additional interventions targeting ventilatory control stability (e.g., acetazolamide or supplemental oxygen) [32-35,62], Our automated approach provides unique advantages by enabling continuous monitoring of LG dynamics across multiple time scales. This Hows quantification of how LG varies with body position [69,70], time of night

[0071] , night-to-night if multi-night data is available, sleep stage, and overall sleep quality at an individual patient level. The method can also track treatment effectiveness over time. While self-similarity mapping provides complementary information about breathing patterns, our LG measurements can possibly stand alone as an independent metric for assessing ventilatory control stability.

[0119] Our findings reveal significant differences in LG across patient groups. Patients with high CAI, elevated SS scores, or heart failure had the highest median LG (0.41) and significantly higher controller gain than the REM-dominant OSA group, emphasizing the prominent role of ventilatory control instability in these populations. Heart failure patients also exhibited the highest median chemosensitivity delay, reflecting chronic respiratory control alterations.

[0120] In contrast, the REM-dominant OSA group exhibited a low median LG of 0.11 with relatively small variability. Nearly no LG values greater than 0.4 were estimated by the EM algorithm for patients with this ventilatory control profile. This finding aligns with our understanding that classic OSA in this group involves stable mechanical obstruction with less influence from ventilatory control instability.

[0121] The NREM-dominant OSA group, likely representing a more heterogeneous mixture of obstructive and central pathologies, showed a significant increase in median LG and higher variability compared to the REM-dominant group. This suggests a complex interplay between physical obstruction and ventilatory control mechanisms, highlighting the necessity for personalized treatment strategies in this subgroup.

[0122] The relationship between altitude and LG was explored using data from healthy patients recorded at various elevations. The results indicated that overnight breathing at sea level is associated with a lower LG compared to higher altitudes. Both average LG and the percentage of sleep with high SS breathing oscillations increased with altitude, consistent with known physiological responses to hypoxia, which include increased chemosensitivity and changes in arousal thresholds. While it is known that the low oxygen at altitude induces sensitization of the hypoxic ventilatory response, acclimatization was also observed in multiple subjects.

[0123] The gradual increase in average LG across categories, despite remaining low (~0.5), underscores the multifactorial nature of ventilatory control instability. While stable breathing periods were excluded, unstable periods vary in morphology and drivers even within the same individual. In subgroups like NREM-dominant OSA, heart failure, and high-altitude apnea, factors such as arousals and airway vulnerabilities influence sleep apnea pathophysiology independently of LG. Periods of SS almost have high loop gain. Similarly, SS varies dynamically and rarely reaches 100% expression across a full sleep period, consistent with the observed median segment duration of 8 minutes. These findings emphasize the complex factors influencing ventilatory control and their importance in personalized sleep apnea management.

[0124] Recent studies emphasize central mechanisms in OSA, suggesting drivedependent OSA, characterized by declining ventilatory drive, accounts for many cases

[0042] , This has significant implications for apnea therapy, highlighting the need to address and stabilize neural ventilatory drive by targeting treatments that either elevate or prevent its decline in addition to mechanical airway patency. Understanding and detecting LG can thus play a critical role in diagnosing and treating drive-dependent OSA.

[0125] To promote patient-level endotype targeted apnea therapy, the proposed method offers a clinically applicable approach to study patient-specific ventilatory LG. While measuring LG using mechanical ventilation has been established for several decades [10,11,48], this method requires dedicated equipment which is typically not available during routing sleep studies. The more recent PUP method that can analyse LG from standard PSG signals made an important step towards the clinical implementation of LG estimation [17,72], however their method requires significant processing and manual labelling of respiratory events prior to analysis. Cardiopulmonary coupling can identifyHLG through detection of narrow-band coupling, which reflects SS patterns in physiological rhythms. However, this approach necessitates 15-17 uninterrupted minutes of consistent cardiopulmonary oscillations to be effective [18,19], making it less sensitive to short bursts of periodic breathing. Limitations of our previous work using SS in RIP data include the method’s high sensitivity in detecting SS oscillations

[0063] , which sometimes leads to the inclusion of noise, especially at lower SS scores. The proposed approach rivals the accuracy of current LG estimation standards while enabling broader implementation by relying solely on RIP signals, eliminating the need for event labelling. This innovation offers opportunities for scalable and precise LG estimation, crucial for personalized OSA therapy. The method could also be adapted for home sleep apnea testing using respiratory effort tracking. However, further validation is needed to extend LG measurements to non-standard signals such as oximetry, jaw movements, or pulse photopl ethy smogram s .Limitations

[0126] Quantifying LG parameters across an entire overnight study requires caution. Even in patients with severe HLG, some segments will show apneas amplified by an underlying increased gain, whereas other parts of the recording, e.g., during REM, can display regular breathing or classic OSA with a normal LG. Stable breathing periods without events can lower whole-night LG averages. Our empirical approach using 8- minute windows provided robust estimates, but LG's variability highlights the need for careful consideration of window size and estimation resolution. Future work should refine and validate these methods across diverse patient populations.

[0127] Occasionally, the model is inaccurate in predicting HLG for individual segments, particularly exhibiting a tendency to underestimate the controller gain parameter, y. This underestimation was also observed in our simulation experiments, suggesting a bias in the model's performance for higher LG values. Nevertheless, this limitation presents limited practical problems when assessing a patient's overall LG across an entire recording. The aggregation of segmental estimates mitigates the impact of occasional underestimation, providing a reliable overall assessment of a patient's ventilatory control stability, as demonstrated in FIGs. 20A-20C.

[0128] The accuracy of this method relies on high-quality respiratory signals. Physicians should interpret LG estimates as guidelines, not absolute values, to identify patients athigher risk of ventilatory instability. These insights could inform therapies targeting ventilatory control stability, such as strengthening respiratory drive [36,42,73] alongside traditional treatments for mechanical airway obstruction, potentially enhancing treatment outcomes through a personalized approach.

[0129] FIG. 26 illustrates flowchart depicting a method 2600 for assessing sleep disorders of a patient, according to embodiments of the present disclosure. It is to be appreciated that method 2600 may not include all operations shown or perform the operations in the order shown.

[0130] Method 2600 begins at step 2602 where signals form a patient is received while the patient is asleep. The signals may be representative of sleep motion of the patient, e.g., by measuring lung expansion and contraction over a period of time as described above with reference to FIG. 1.

[0131] At step 2604, ventilation control parameters from the received signals may be extracted. For example, the extraction of the ventilation control parameters may be done based on a Mackey-Glass relationship.

[0132] At step 2606, a loop gain is determined, where the loop gain is a ratio of a first ventilation change to a second ventilation change. The first ventilation change occurs following an end of a disturbance. The second ventilation change is induced by the disturbance.

[0133] At step 2608, a clinical outcome of the patient is predicted based on a higher value or a lower value of the loop gain determined in step 2606.Conclusions

[0134] This study advances the understanding of ventilatory control dynamics in sleep apnea by providing a novel, automated method to estimate LG and its stability parameters. Our findings underscore the importance of personalized approaches to apnea management, considering the complex interplay between mechanical and neural factors in respiratory control. Further research and clinical validation are essential to fully integrate these methods into routine practice, ultimately improving patient outcomes through tailored therapy strategies.SUPPLEMENTAL MATERIAL

[0135] The supplemental material expands on the mathematical principles and computational models presented in the manuscript, particularly focusing on the application of the Mackey-Glass equations for modelling the dynamic behaviour of ventilatory control in sleep apnea patients. The methods detailed here represent a significant advancement in estimating loop gain (LG) by utilizing respiratory inductance plethysmography (RIP) signals, enabling the measurement of ventilatory stability in a non-invasive, scalable manner without relying on traditional, more invasive methods.

[0136] Central to this approach was the formulation of ventilatory control as a closed- loop system, where the body's response to respiratory disturbances — such as variations in CO2 levels and arousal mechanisms — are dynamically regulated. By adopting a recursive Bayesian filter and a fixed-interval smoothing algorithm, the approach accurately estimated hidden physiological parameters like controller gain (y) and chemosensitivity delay (T), which played a pivotal role in assessing LG and overall ventilatory stability. These parameters, while unobservable directly from RIP signals, were inferred through a mathematical model that captured the feedback-driven nature of ventilatory control. This process ultimately supported more precise, personalized management strategies for sleep apnea.

[0137] First, this document describes our assumption-based representation of the respiratory feedback loop, which was translated into the Mackey-Glass equations to define the oscillatory patterns seen in ventilation under different apnea phenotypes. We then introduce the use of the expectation-maximization (EM) algorithm, an iterative technique that optimized the estimation of hidden parameters by maximizing the likelihood of the observed data. Each iteration of the EM algorithm involved an E-step, which calculated the expected log-likelihood of the observed data conditioned on current model parameters, and an M-step, which refined those parameters to maximize the likelihood function. This iterative process continued until convergence, allowing for precise estimations of controller gain, delay, and other critical parameters that governed the stability of respiratory patterns.

[0138] Next, we discuss the Gaussian approximation of the posterior distributions during each filtering step. By assuming that the posterior distribution of each time-step's parameters was Gaussian, the derivation yielded closed-form solutions for the recursivefilter update equations, enhancing computational efficiency. This Gaussian assumption was foundational to the filter’s one-step prediction updates, which enabled real-time estimation of ventilatory stability as breathing oscillated across various stages of sleep or altitude conditions. For improved parameter stability, the filtering and smoothing operations relied on first and second-order derivatives of the likelihood functions, ensuring robust and unbiased updates to the model.

[0139] Finally, this material provides calculations for each component of the observation likelihood function within the Bayesian framework. We also cover the development of a fixed-interval smoother, which leveraged the one-step predictions across the full sequence of respiratory data, refining estimates by incorporating subsequent observations. This approach captured the time-dependent fluctuations in LG and enabled a more reliable analysis of transient high-loop-gain (HLG) segments.

[0140] The autonomously estimated LG parameter represents a scalable and robust tool for clinicians, enabling the analysis of complex respiratory control dynamics in sleep apnea and making it adaptable to various apnea endotypes, altitude conditions, and clinical settings.MACKEY GLASS EQUATIONS

[0141] The Mackey-Glass equation was our fundamental function for modelling the dynamic behaviour of ventilatory control in sleep apnea. Its general form below describes how arterial CO2 concentration (X) is influenced by various parameters within the respiratory control system. whereParameter Physiological Unit Physiological representation rangeL CO2 production rate % per sec 0.05Vm Maximum ventilation ‘scaled’ 1 rateX Arterial CO2 0 - 100 concentration y Gain ‘unitless’ 0.1 - 1.2T Chemosensitivity time sec 5 - 50 delay

[0142] Building upon this we implemented a parameter for breathing disturbance (U) combined with a scaling parameter corresponding to the reduction in ventilation during respiratory events, and arousal (var), as constant transient 5 second increases in ventilation. Resulting in the following:

[0143] FIG. 25 provides a visual representation of both observed and modelled aspects of ventilation control. It displays segments of observed breathing, upper airway opening, and the corresponding modelled levels of CO2 and ventilation over time, when introduced to breathing events.MODELLING MATHEMATICS

[0144] The model was built on several key assumptions and parameters to simplify and focus the calculation process. For instance, the maximum ventilation rate (Vmax) was set to 1, the scaling factor (0) to 1, and the duration of each arousal event (w) to 5 seconds. Additionally, the breathing disturbance (uk) was derived from surrogate ventilation, allowing for a realistic representation of respiratory events in the absence of direct measurements.

[0145] The observation process was defined by the following equation:where skis unit step function:

[0146] This step function, combined with the arousal effectenabled the model to simulate the impact of transient ventilation increases due toarousals. The observation noise (i9k) was assumed to follow a normal distribution with mean 0 and variancecapturing random variations in observed ventilation.

[0147] The hidden state of the system, representing the unobserved arterial CO2 concentration, was modelled as:where £kis the process noise, normally distributed with mean 0 and variance aj. The free parameters of the model, collectively represented as included thechemosensitivity delay (T), the CO2 production rate (L), the gain (y), and the arousal scaling factor (h). These parameters allowed the model to adapt to individual respiratory control dynamics, enabling accurate predictions of ventilatory stability.STATE PREDICTION ALGORITHM

[0148] We assumed that the model parameters are known and therefore used the recursive Bayes filter and smoother to calculate the posterior estimate of xkgiven full observation Here, we focus on a Gaussian approximation technique to derive aclosed-form solution for the filter estimation of the state variables. In the following subsection the filter estimation of the state variables is described.Filter Estimation:

[0149] To build the approximation solution, we assumed that the posterior distribution over xkat each time interval can be approximated by a multivariate Gaussian distribution. When we define the mean and covariance of the filter estimation at time k as respectively, it results in the following one-step prediction update rules:

[0150] Using the likelihood function at time k andthe one step prediction, the filter update was defined as:whereis the observation process likelihood function.defined the filter and one-step probability distribution of xk.

[0151] To find the update rule for xk\kand a^k, we took the logarithm on both sides and computed the first and second-order derivative. We then found the update rule by setting\1stderivation:

[0152] The final form for the mean of the filter solution was defined as:2ndderivation:

[0153] The final form for the covariance of the filter solution was defined as:Fixed Interval Smoother:

[0154] Given the sequence of posterior one-step estimates, we used the fixed-interval smoothing algorithm to compute . The smoothing algorithm was:for k = K — 1, ... ,1 and initial conditionMODEL PARAMETER ESTIMATION

[0155] The free parameters of the model included the state transition process parameters and the observation process free parameter, i.e., {T, L, y, h}. We utilized the expectationmaximization (EM) algorithm to estimate all unknown parameters. The EM algorithm is an iterative optimization process that finds the maximum likelihood estimate of the model parameters by recursively applying two computations, i.e., E-step, M-step.E-step:

[0156] For the E-step, at iteration (Z + 1), we calculated the expectation of the complete log-likelihood as defined by the joint probability distribution of the state vector x0 Kand full observation V° ,K, given the model estimated parameters from the previous iteration (Z). The complete log-likelihood function is defined by:

[0157] P(xo|0) has a normal distribution, independent from the free parameters. Therefore, we rewrote Equation this using a constant, C4. logP("4

[0158] Finally, we calculated the following expectation function:

[0159] To find the expectation values calculated the following terms:

[0160] To solve these two terms, we first applied a Taylor Expansion:Final form:Similarly, to calculateM-step:

[0161] For the M-step, we estimated a new set of parameters that maximize the expected log-likelihood function defined in Equation 11. The new set of parameters for iteration (Z + 1) were computed by maximizing Q®. The maximization step was defined by:arg max Q®<P

[0162] This maximization process was done analytically when there is a closed form solution for and its derivative with respect to the model free parameters. The otherpossible solution for the parameter update is numerical methods like gradient ascent method.

[0163] The EM algorithm ran multiple iterations until it converges to a local maximum / best fit. We terminated the EM iterations when Q slows its increment below threshold, e:

[0164] For example: whereCOMPUTER EMBODIMENTS

[0165] The figures as described herein are illustrative examples allowing an explanation of the present disclosure. It should be understood that embodiments of the present disclosure could be implemented in hardware, firmware, software, or a combination thereof. In such an embodiment, the various components and steps would be implemented in hardware, firmware, and / or software to perform the functions of the present disclosure. That is, the same piece of hardware, firmware, or module of software could perform one or more of the illustrated blocks (e.g., components or steps).

[0166] The present disclosure may be implemented in one or more computer systems capable of carrying out the functionality described herein. Referring to FIG. 14, an example computer system 1400 useful in implementing the present disclosure is shown. Various embodiments of the invention are described in terms of this example computer system 1400. After reading this description, it will become apparent to one skilled in the relevant art(s) how to implement the invention using other computer systems and / or computer architectures.

[0167] The computer system 1400 includes one or more processors, such as processor 1404. The processor 1404 is connected to a communication infrastructure 1406 (e.g., a communications bus, crossover bar, or network).

[0168] Computer system 1400 may include a display interface 1402 that forwards graphics, text, and other data from the communication infrastructure 1406 (or from a frame buffer not shown) for display on the display unit 1430.

[0169] Computer system 1400 also includes a main memory 1408, preferably random access memory (RAM), and may also include a secondary memory 1410. The secondary memory 1410 may include, for example, a hard disk drive 1412 and / or a removable storage drive 1414, representing a floppy disk drive, a magnetic tape drive, an optical disk drive, etc. The removable storage drive 1414 reads from and / or writes to a removable storage unit 1418 in a well-known manner. Removable storage unit 1418, represents a floppy disk, magnetic tape, optical disk, etc. which is read by and written to removable storage drive 1414. As will be appreciated, the removable storage unit 1418 includes a computer usable storage medium having stored therein computer software (e.g., programs or other instructions) and / or data.

[0170] In alternative embodiments, secondary memory 1410 may include other similar means for allowing computer software and / or data to be loaded into computer system 1400. Such means may include, for example, a removable storage unit 1422 and an interface 1420. Examples of such may include a program cartridge and cartridge interface (such as that found in video game devices), a removable memory chip (such as an EPROM, or PROM) and associated socket, and other removable storage units 1422 and interfaces 1420 which allow software and data to be transferred from the removable storage unit 1422 to computer system 1400.

[0171] Computer system 1400 may also include a communications interface 1424. Communications interface 1424 allows software and data to be transferred between computer system 1400 and external devices. Examples of communications interface 1424 may include a modem, a network interface (such as an Ethernet card), a communications port, a PCMCIA slot and card, etc. Software and data transferred via communications interface 1424 are in the form of signals 1428 which may be electronic, electromagnetic, optical, or other signals capable of being received by communications interface 1424. These signals 1428 are provided to communications interface 1424 via a communications path (e.g., channel) 1426. Communications path 1426 carries signals 1428 and may be implemented using wire or cable, fiber optics, a phone line, a cellular phone link, an RF link, free-space optics, and / or other communications channels.

[0172] In this document, the terms “computer program medium” and “computer usable medium” are used to generally refer to media such as removable storage unit 1418, removable storage unit 1422, a hard disk installed in hard disk drive 1412, and signals 1428. These computer program products are means for providing software to computer system 1400. The invention is directed to such computer program products.

[0173] Computer programs (also called computer control logic or computer readable program code) are stored in main memory 1408 and / or secondary memory 1410. Computer programs may also be received via communications interface 1424. Such computer programs, when executed, enable the computer system 1400 to implement the present disclosure as discussed herein. In particular, the computer programs, when executed, enable the processor 1404 to implement the processes of the present disclosure described above. Accordingly, such computer programs represent controllers of the computer system 1400.

[0174] In an embodiment where the invention is implemented using software, the software may be stored in a computer program product and loaded into computer system 1400 using removable storage drive 1414, hard disk drive 1412, interface 1420, or communications interface 1424. The control logic (software), when executed by the processor 1404, causes the processor 1404 to perform the functions of the invention as described herein.

[0175] In another embodiment, the invention is implemented primarily in hardware using, for example, hardware components such as application specific integrated circuits (ASICs). Implementation of the hardware state machine so as to perform the functions described herein will be apparent to one skilled in the relevant art(s).

[0176] In yet another embodiment, the invention is implemented using a combination of both hardware and software.

[0177] In one example embodiment, the present disclosure may be implemented in a computer-based monitor unit for use in a clinical setting. In another embodiment, the present disclosure may be implemented in an ambulatory unit akin to a Holter monitor, personal computing device, or similar portable device. In yet another embodiment, the present disclosure may be implemented in an implantable medical device such as an implantable cardioverter defibrillator (ICD).

[0178] The foregoing description of the specific embodiments will so fully reveal the general nature of the invention that others can, by applying knowledge within the skill of the art (including the contents of the documents cited and incorporated by reference herein), readily modify and / or adapt for various applications such specific embodiments, without undue experimentation, without departing from the general concept of the present disclosure. Therefore, such adaptations and modifications are intended to be within the meaning and range of equivalents of the disclosed embodiments, based on the teaching and guidance presented herein. It is to be understood that the phraseology or terminology herein is for the purpose of description and not of limitation, such that the terminology or phraseology of the present specification is to be interpreted by the skilled artisan in light of the teachings and guidance presented herein, in combination with the knowledge of one skilled in the art.

[0179] While various embodiments of the present disclosure have been described above, it should be understood that they have been presented by way of example, and not limitation. It will be apparent to one skilled in the relevant art(s) that various changes in form and detail can be made therein without departing from the spirit and scope of the invention. Thus, the present disclosure should not be limited by any of the abovedescribed exemplary embodiments, but should be defined only in accordance with the following claims and their equivalents.REFERENCES1. Zhao X, Li X, Xu H, Qian Y, Fang F, Yi H, Guan J, Yin SK. Relationships between cardiometabolic disorders and obstructive sleep apnea: Implications for cardiovascular disease risk. J Clin Hypertens (Greenwich) 2019; 21 : 280-290.2. Lajoie AC, Lafontaine AL, Kimoff RJ, Kaminska M. Obstructive Sleep Apnea in Neurodegenerative Disorders: Current Evidence in Support of Benefit from Sleep Apnea Treatment. J Clin Med 2020; 9.3. Deacon NL, Catcheside PG. The role of high loop gain induced by intermittent hypoxia in the pathophysiology of obstructive sleep apnoea. Sleep Med Rev 2015; 22: 3-14.4. Lee JJ, Sundar KM. Evaluation and Management of Adults with Obstructive Sleep Apnea Syndrome. Lung 2021; 199: 87-101.5. McClean PA, Phillipson EA, Martinez D, Zamel N. Single breath of CO2 as a clinical test of the peripheral chemoreflex. J Appl Physiol (1985) 1988; 64: 84-89.Ghazanshahi SD, Khoo MC. Estimation of chemoreflex loop gain using pseudorandom binary CO2 stimulation. IEEE Trans Biomed Eng 1997; 44: 357-366. Hudgel DW, Gordon EA, Thanakitcharu S, Bruce EN. Instability of ventilatory control in patients with obstructive sleep apnea. Am J Respir Crit Care Med 1998; 158: 1142-1149. Younes M, Ostrowski M, Atkar R, Laprairie J, Siemens A, Hanly P. Mechanisms of breathing instability in patients with obstructive sleep apnea. J Appl Physiol (1985) 2007; 103: 1929-1941. Loewen A, Ostrowski M, Laprairie J, Atkar R, Gnitecki J, Hanly P, Younes M. Determinants of ventilator instability in obstructive sleep apnea: inherent or acquired? Sleep 2009; 32: 1355-1365. Meza S, Younes M. Ventilatory stability during sleep studied with proportional assist ventilation (P AV). Sleep 1996; 19: S164-166. Younes M, Ostrowski M, Thompson W, Leslie C, Shewchuk W. Chemical control stability in patients with obstructive sleep apnea. Am J Respir Crit Care Med 2001; 163: 1181-1190. Stanley NN, Cunningham EL, Altose MD, Kelsen SG, Levinson RS, Chemiack NS. Evaluation of breath holding in hypercapnia as a simple clinical test of respiratory chemosensitivity. Thorax 1975; 30: 337- 343. Trembach N, Zabolotskikh I. Breath-holding test in evaluation of peripheral chemoreflex sensitivity in healthy subjects. Respir Physiol Neurobiol 2017; 235: 79-82. Messineo L, Taranto-Montemurro L, Azarbarzin A, Oliveira Marques MD, Calianese N, White DP, Wellman A, Sands SA. Breath-holding as a means to estimate the loop gain contribution to obstructive sleep apnoea. J Physiol 2018; 596: 4043-4056. Terrill PI, Edwards BA, Nemati S, Butler JP, Owens RL, Eckert DJ, White DP, Malhotra A, Wellman A, Sands SA. Quantifying the ventilatory control contribution to sleep apnoea using polysomnography. Eur Respir J 2015; 45: 408-418. Nava-Guerra L, Edwards BA, Terrill PI, Sands SA, Amin RS, Kemp JS, Khoo MCK. Quantifying ventilatory control stability from spontaneous sigh responses during sleep: a comparison of two approaches. Physiol Meas 2018; 39: 114005. Finnsson E, Olafsdottir GH, Loftsdottir DL, Jonsson SAE, Helgadottir H, Agustsson JS, Sands SA, Wellman A. A scalable method of determining physiological endotypes of sleep apnea from a polysomnographic sleep study. Sleep 2021; 44.Ramar K, Desrues B, Ramar P, Morgenthaler TI. Analysis of cardiopulmonary coupling to assess adaptive servo-ventilation success in complex sleep apnea management. Sleep Breath 2013; 17: 861-866. Al Ashry HS, Ni Y, Thomas RJ. Cardiopulmonary Sleep Spectrograms Open a Novel Window Into Sleep Biology -Implications for Health and Disease. Front Neurosci 2021; 15: 755464. Oppersma E, Ganglberger W, Sun H, Thomas RJ, Westover MB. Algorithm for automatic detection of self- similarity and prediction of residual central respiratory events during continuous positive airway pressure. Sleep 2021; 44. Rao H, Thomas RJ. Complex sleep apnea. Curr Treat Options Neurol 2013; 15: 677-691. Ginter G, Sankari A, Eshraghi M, Obiakor H, Yarandi H, Chowdhuri S, Salloum A, Badr MS. Effect of acetazolamide on susceptibility to central sleep apnea in chronic spinal cord injury. J Appl Physiol (1985) 2020; 128: 960-966. Sands SA, Edwards BA, Kee K, Turton A, Skuza EM, Roebuck T, O'Driscoll DM, Hamilton GS, Naughton MT, Berger PJ. Loop gain as a means to predict a positive airway pressure suppression of Cheyne-Stokes respiration in patients with heart failure. Am J Respir Crit Care Med 2011; 184: 1067-1075. Edwards BA, Andara C, Landry S, Sands SA, Joosten SA, Owens RL, White DP, Hamilton GS, Wellman A. Upper- Airway Collapsibility and Loop Gain Predict the Response to Oral Appliance Therapy in Patients with Obstructive Sleep Apnea. Am J Respir Crit Care Med 2016; 194: 1413-1422. Deacon-Diaz NL, Sands SA, McEvoy RD, Catcheside PG. Daytime loop gain is elevated in obstructive sleep apnea but not reduced by CPAP treatment. J Appl Physiol (1985) 2018; 125: 1490-1497. Op de Beeck S, Wellman A, Dieltjens M, Strohl KP, Willemen M, Van de Heyning PH, Verbraecken JA, Vanderveken OM, Sands SA, Investigators ST. Endotypic Mechanisms of Successful Hypoglossal Nerve Stimulation for Obstructive Sleep Apnea. Am J Respir Crit Care Med 2021; 203: 746-755. Wong AM, Landry SA, Joosten SA, Thomson LDJ, Turton A, Stonehouse J, Mansfield DR, Burgess G, Hays A, Sands SA, Andara C, Beatty CJ, Hamilton GS, Edwards BA. Examining the impact of multilevel upper airway surgery on the obstructive sleep apnoeaendotypes and their utility in predicting surgical outcomes. Respirology 2022; 27: SOO- 899. Mulgrew AT, Lawati NA, Ayas NT, Fox N, Hamilton P, Cortes L, Ryan CF. Residual sleep apnea on polysomnography after 3 months of CPAP therapy: clinical implications, predictors and patterns. Sleep Med 2010; 11 : 110-125. Denotti AL, Wong KK, Dungan GC, Gilholme JW, Marshall NS, Grunstein RR. Residual sleep-disordered breathing during autotitrating continuous positive airway pressure therapy. Eur Respir .72012; 30: 1301- 1307. Reiter J, Zleik B, Bazalakova M, Mehta P, Thomas RJ. Residual Events during Use of CPAP: Prevalence, Predictors, and Detection Accuracy. J Clin Sleep Med 2016; 12: 1153-1158. Sands SA, Edwards BA, Terrill PI, Butler JP, Owens RL, Taranto-Montemurro L, Azarbarzin A, Marques M, Hess LB, Smales ET, de Melo CM, White DP, Malhotra A, Wellman A. Identifying obstructive sleep apnoea patients responsive to supplemental oxygen therapy. Eur Respir J 2018; 52. Mulchrone A, Shokoueinejad M, Webster J. A review of preventing central sleep apnea by inspired CO2. Physiol Meas 2016; 37: R36-45. Edwards BA, Sands SA, Eckert DJ, White DP, Butler JP, Owens RL, Malhotra A, Wellman A. Acetazolamide improves loop gain but not the other physiological traits causing obstructive sleep apnoea. J Physiol 2012; 590: 1199-1211. Eskandari D, Zou D, Karimi M, Stenlof K, Grote L, Hedner J. Zonisamide reduces obstructive sleep apnoea: a randomised placebo-controlled study. Eur Respir J 2014; 44: 140-149. Schmickl CN, Landry S, Orr JE, Nokes B, Edwards BA, Malhotra A, Owens RL. Effects of acetazolamide on control of breathing in sleep apnea patients: Mechanistic insights using meta-analyses and physiological model simulations. Physiol Rep 2021; 9: el5071. Hedner J, Stenlof K, Zou D, Hoff E, Hansen C, Kuhn K, Lennartz P, Grote L. A Randomized Controlled Clinical Trial Exploring Safety and Tolerability of Sulthiame in Sleep Apnea. Am J Respir Crit Care Med 2022; 205: 1461-1469. Ni YN, Thomas RJ. Predictors and consequences of residual apnea during positive airway pressure therapy. Sleep Med 2023; 106: 42-51.Ni YN, Holzer RC, Thomas RJ. Acute and long-term effects of acetazolamide in presumed high loop gain sleep apnea. Sleep Med 2023; 107: 137-148. Troester MMQ, S.F.; Berry, R.B., et al. The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications. American Academy of Sleep Medicine; 2023. Nassi TE. Algorithms for Automated Scoring of Respiratory Events in Sleep. Enschede, Netherlands: University of Twente; 2021. Thomas RJ, Terzano MG, Parrino L, Weiss JW. Obstructive sleep-disordered breathing with a dominant cyclic alternating pattern— a recognizable polysomnographic variant with practical clinical implications. Sleep 2004; 27: 229-234. Gell LK, Vena D, Alex RM, Azarbarzin A, Calianese N, Hess LB, Taranto-Montemurro L, White DP, Wellman A, Sands SA. Neural ventilatory drive decline as a predominant mechanism of obstructive sleep apnoea events. Thorax 2022; 77: 707-716. Sankri-Tarbichi AG, Rowley JA, Badr MS. Expiratory pharyngeal narrowing during central hypocapnic hypopnea. Am J Respir Crit Care Med 2009; 179: 313-319. Badr MS, Toiber F, Skatrud JB, Dempsey J. Pharyngeal narrowing / occlusion during central sleep apnea. J Appl Physiol (1985) 1995; 78: 1806-1815. Thomas RJ, Tamisier R, Boucher J, Kotlar Y, Vigneault K, Weiss JW, Gilmartin G. Nocturnal hypoxia exposure with simulated altitude for 14 days does not significantly alter working memory or vigilance in humans. Sleep 2007; 30: 1195-1203. Lindsey BG, Nuding SC, Segers LS, Morris KF. Carotid Bodies and the Integrated Cardiorespiratory Response to Hypoxia. Physiology (Bethesda) 2018; 33: 281-297. Eckert DJ, Jordan AS, Merchia P, Malhotra A. Central sleep apnea: Pathophysiology and treatment. Chest 2007,' 131 : 595-607. Dempsey JA. Central sleep apnea: misunderstood and mistreated! F1000Res 2019; 8. Wellman A, Eckert DJ, Jordan AS, Edwards BA, Passaglia CL, Jackson AC, Gautam S, Owens RL, Malhotra A, White DP. A method for measuring and modeling the physiological traits causing obstructive sleep apnea. J Appl Physiol (1985) 2011; 110: 1627-1637. Stoberl AS, Schwarz El, Haile SR, Turnbull CD, Rossi VA, Stradling JR, Kohler M. Night-to-night variability of obstructive sleep apnea. J Sleep Res 2017; 26: 782-788.Cunha G, Apostolo A, De Martino F, Salvioni E, Matavelli I, Agostoni P. Exercise oscillatory ventilation: the past, present, and future. Eur J Prev Cardiol 2023;30(Suppl 2):ii22-ii27. Goulart CDL, Silva RN, Agostoni P, Franssen FME, Myers J, Arena R, Borghi-Silva A. Exercise oscillatory ventilation in patients with coexisting chronic obstructive pulmonary disease and heart failure: Clinical implications. Respir Med 2023 ;217: 107332. doi:10.1016 / j.rmed.2023.107332. Gama F, Rocha B, Aguiar C, Strong C, Freitas P, Brizido C, Tralhao A, Durazzo A, Mendes M. Exercise Oscillatory Ventilation Improves Heart Failure Prognostic Scores. Heart Lung Circ 2023;32:949-957. Cunha GJL, Maltes S, Rocha BML, Nina D, Aguiar C, Andrade MJ, Moreno L, Durazzo A, Mendes M, Agostoni P. Beyond exercise oscillatory ventilations: the prognostic impact of loop gain in heart failure. Eur J Prev Cardiol 2023:zwad021. doi: 10.1093 / eurjpc / zwad021. Gentile F, Fabiani I, Emdin M. Exercise oscillatory ventilation in heart failure and brain- lung-heart-muscle crosstalk. Eur J Prev Cardiol 2022;;29: 1689-1691. Cassady SJ, Glick DR, Williams SR, Lewis GD. Periodic Breathing in Cardiopulmonary Exercise Testing: Exercise Oscillatory Ventilation in Heart Failure. Ann Am Thorac Soc 2022;19: 120-126. Bjornsdottir E, Keenan BT, Eysteinsdottir B, et al. Quality of life among untreated sleep apnea patients compared to the general population and changes after treatment with positive airway pressure. Journal of sleep research. 2015; 24 (3). Ramirez J-M, Alfredo J. Garcia I, Anderson TM, et al. Central and Peripheral factors contributing to Obstructive Sleep Apneas. Respiratory physiology & neurobiology. 2013; 189 (2). Zinchuk A, Gentry M, Concato J, Yaggi K. Phenotypes in obstructive sleep apnea: a definition, examples and evolution of approaches. Sleep medicine reviews. 2017; 35. Wellman A, Jordan AS, Malhotra A, et al. Ventilatory Control and Airway Anatomy in Obstructive Sleep Apnea. American journal of respiratory and critical care medicine. 2004; 170 (11). Orr JE, Malhotra A, Sands SA. Pathogenesis of Central and Complex Sleep Apnoea. Respirology. 2017; 22 (1).Sands SA, Edwards BA, Terrill PI, et al. Identifying obstructive sleep apnoea patients responsive to supplemental oxygen therapy. Eur Respir J. 2018; 52 (3). Nassi TE, Oppersma E, Labarca G, Donker DW, Westover MB, Thomas RJ. Morphological Prediction of CPAP Associated Acute Respiratory Instability. Annals of the American Thoracic Society. 2024. Redeker NS, Jeon S, Muench U, Campbell D, Walsleben J, Rapoport DM. Insomnia Symptoms and Daytime Function in Stable Heart Failure. Sleep. 2010; 33 (9). Redeker NS, Muench U, Zucker MJ, et al. Sleep Disordered Breathing, Daytime Symptoms, and Functional Performance in Stable Heart Failure. Sleep. 2010; 33 (4). Landry SA, Andara C, Terrill PI, et al. Ventilatory control sensitivity in patients with obstructive sleep apnea is sleep stage dependent. Sleep. 2018; 41 (5). Truong C, Oudre L, Vayatis N. Selective review of offline change point detection methods. Signal Processing. 2020; 167. Do CB, Batzoglou S, Do CB, Batzoglou S. What is the expectation maximization algorithm? Nature Biotechnology. 2008; 26 (8). Joosten SA, Landry SA, Sands SA, et al. Dynamic loop gain increases upon adopting the supine body position during sleep in patients with obstructive sleep apnoea. Respirology. 2017; 22 (8). Messineo L, Joosten S, Perger E. Mechanisms relating to sleeping position to the endotypes of sleep disordered breathing. Current Opinion in Pulmonary Medicine. 2023;29 (6). Tkacova R, Niroumand M, Lorenzi -Filho G, Bradley TD. Overnight Shift From Obstructive to Central Apneas in Patients With Heart Failure. Circulation. 2001; 103 (2). Orr JE, Sands SA, Edwards BA, et al. Measuring Loop Gain via Home Sleep Testing in Patients with Obstructive Sleep Apnea. American Journal of Respiratory and Critical Care Medicine. 2018; 197 (10). Quinn T, Thomas RJ, Heckman EJ. Enhanced expiratory rebreathing space for high loop gain sleep apnea treatment. Frontiers in Sleep. 2023; 2.

Claims

WHAT IS CLAIMED IS:

1. A method of assessing sleep disorders of a patient, the method comprising: receiving signals from the patient over a period of time while the patient is asleep, the signals being representative of sleep motion of the patient; identifying a complex within the received signals, the identifying including ascertaining an envelope of the received signals over the period of time; determining a ratio of a maximum envelope height to a minimum envelope height within the complex; determining a duration between a maximum envelope height of the complex and a subsequent maximum envelope of a subsequent complex; computing a convolution of a positive envelope and a rotated negative envelope of the complex, the rotated negative envelope being a rotation of the negative envelope about a time axis; assessing a time location of a maximum distance between the positive envelope and the negative envelope; and determining a self-similarity of the received signals based on the identifying, determining, determining, computing, and assessing.

2. The method of claim 1, further comprising determining a first duration between the minimum envelope height and the maximum envelope height and a second duration between the maximum envelope height and subsequent minimum envelope height.

3. The method of claim 2, further comprising determining a ratio of a first value to a second value, the first value being a subtraction of the first duration and second duration, and the second value being an addition of the first and second duration.

4. The method of claim 1, further comprising comparing the ratio with a threshold value, the threshold value being 0.15.

5. The method of claim 1, further comprising comparing the convolution with a threshold value, the threshold value being 0.7.

6. The method of claim 1, wherein the period of time exceeds 10 seconds.

7. The method of claim 1, further comprising determining a likelihood of failure of continuous positive airway pressure (CPAP) therapy based on the determined selfsimilarity.

8. A sleep disorder assessment system, the system comprising: an input module configured to receive signals from the patient over a period of time while the patient is asleep, the signals being representative of sleep motion of the patient; and a processor configured to: receive signals from the patient over a period of time while the patient is asleep, the signals being representative of sleep motion of the patient; identify a complex within the received signals, the identifying including ascertaining an envelope of the received signals over the period of time; determine a ratio of a maximum envelope height to a minimum envelope height within the complex; determine a duration between a maximum envelope height of the complex and a subsequent maximum envelope of a subsequent complex; compute a convolution of a positive envelope and a rotated negative envelope of the complex, the rotated negative envelope being a rotation of the negative envelope about a time axis; assess a time location of a maximum distance between the positive envelope and the negative envelope; and determine a self-similarity of the received signals based on the identifying, determining, determining, computing, and assessing.

9. The system of claim 8, wherein the processor is further configured to determine a first duration between the minimum envelope height and the maximum envelope height and a second duration between the maximum envelope height and subsequent minimum envelope height.

10. The system of claim 9, wherein the processor is further configured to determine a ratio of a first value to a second value, the first value being a subtraction of the first duration and second duration, and the second value being an addition of the first and second duration.

11. The system of claim 8, wherein the processor is further configured to compare the ratio with a threshold value, the threshold value being 0.15.

12. The system of claim 8, wherein the processor is further configured to compare the convolution with a threshold value, the threshold value being 0.7.

13. The system of claim 8, wherein the period of time exceeds 10 seconds.

14. The system of claim 8, wherein the processor is further configured to determine a likelihood of failure of continuous positive airway pressure (CPAP) therapy based on the determined self-similarity.

15. A computer program product stored on a computer readable media, including a set of instructions that, when executed by a computing device, perform a method of assessing sleep disorders of a patient, comprising: receiving signals from the patient over a period of time while the patient is asleep, the signals being representative of sleep motion of the patient; identifying a complex within the received signals, the identifying including ascertaining an envelope of the received signals over the period of time; determining a ratio of a maximum envelope height to a minimum envelope height within the complex; determining a duration between a maximum envelope height of the complex and a subsequent maximum envelope of a subsequent complex; computing a convolution of a positive envelope and a rotated negative envelope of the complex, the rotated negative envelope being a rotation of the negative envelope about a time axis; assessing a time location of a maximum distance between the positive envelope and the negative envelope; anddetermining a self-similarity of the received signals based on the identifying, determining, determining, computing, and assessing.

16. The computer program product of claim 15, wherein the method further comprises determining a first duration between the minimum envelope height and the maximum envelope height and a second duration between the maximum envelope height and subsequent minimum envelope height.

17. The computer program product of claim 16, wherein the method further comprises determining a ratio of a first value to a second value, the first value being a subtraction of the first duration and second duration, and the second value being an addition of the first and second duration.

18. The computer program product of claim 15, wherein the method further comprises comparing the ratio with a threshold value, the threshold value being 0.15.

19. The computer program product of claim 15, wherein the method further comprises comparing the convolution with a threshold value, the threshold value being 0.7.

20. The computer program product of claim 13, wherein the period of time exceeds 10 seconds.

21. A method of assessing sleep disorders of a patient, the method comprising: receiving signals from the patient over a period of time while the patient is asleep, the signals being representative of sleep motion of the patient; extracting ventilation control parameters from the received signals based on a Mackey-Glass relationship; determining a loop gain, the loop gain being a ratio of a first ventilation change following an end of a disturbance to a second ventilation change induced by the disturbance; and predicting a clinical outcome of the patient based on a higher value or a lower value of the loop gain.

22. The method of claim 21, wherein the signals are abdominal respiratory inductance plethysmography (RIP) signals.

23. The method of claim 21, wherein the period of time is 8 minutes.

24. The method of claim 21, wherein the period of time is one night.

25. The method of claim 21, wherein the ventilation control parameters include a delay between lungs of the patient and brainstem of the patient.

26. The method of claim 21, wherein the ventilation control parameters include a ventilation rate.

27. The method of claim 21, wherein the disturbance includes an apnea or a hypopnea.

28. A sleep disorder assessment system, the system comprising: an input module configured to receive signals from the patient over a period of time while the patient is asleep, the signals being representative of sleep motion of the patient; and a processor configured to: extract ventilation control parameters from the received signals based on a Mackey-Glass relationship; determine a loop gain, the loop gain being a ratio of a first ventilation change following an end of a disturbance to a second ventilation change induced by the disturbance; and predict a clinical outcome of the patient based on a higher value or a lower value of the loop gain.

29. The sleep disorder assessment system of claim 28, wherein the signals are abdominal respiratory inductance plethysmography (RIP) signals.

30. The sleep disorder assessment system of claim 28, wherein the period of time is 8 minutes.

31. The sleep disorder assessment system of claim 28, wherein the period of time is one night.

32. The sleep disorder assessment system of claim 28, wherein the ventilation control parameters include a delay between lungs of the patient and brainstem of the patient.

33. The sleep disorder assessment system of claim 28, wherein the ventilation control parameters include a ventilation rate.

34. The sleep disorder assessment system of claim 28, wherein the disturbance includes an apnea or a hypopnea.

35. A computer program product stored on a computer readable media, including a set of instructions that, when executed by a computing device, perform a method of assessing sleep disorders of a patient, comprising: receiving signals from the patient over a period of time while the patient is asleep, the signals being representative of sleep motion of the patient; extracting ventilation control parameters from the received signals based on a Mackey-Glass relationship; determining a loop gain, the loop gain being a ratio of a first ventilation change following an end of a disturbance to a second ventilation change induced by the disturbance; and predicting a clinical outcome of the patient based on a higher value or a lower value of the loop gain.

36. The computer program product of claim 35, wherein the signals are abdominal respiratory inductance plethysmography (RIP) signals.

37. The computer program product of claim 35, wherein the period of time is 8 minutes.

38. The computer program product of claim 35, wherein the period of time is one night.

39. The computer program product of claim 35, wherein the ventilation control parameters include a delay between lungs of the patient and brainstem of the patient.

40. The computer program product of claim 35, wherein the ventilation control parameters include a ventilation rate.

41. The computer program product of claim 35, wherein the disturbance includes an apnea or a hypopnea.

Images