Con: can physiological risk factors for obstructive sleep apnea be determined by analysis of data obtained from routine polysomnography?

The following argument was prepared in response to the question without the knowledge of the contents of the opposing argument. The answer is CLEARLY NO. The mechanisms of ventilatory instability in obstructive sleep apnea (OSA) are complex [1, 2], and defining them with confidence requires interventions that are not practical in routine polysomnography [3–5]. If it is not possible to measure instability traits from routine polysomnograms (PSGs), how is it that since 2015 papers and reviews are being published at an impressive rate in top-rated journals using an approach that claims to estimate all the stability factors from standard clinical PSGs? All these papers are based on a method proposed by Terrill et al. in 2015 to measure loop gain (LG) [6], and subsequently “improved” to provide more and more traits [7]. The basic approach is to measure ventilation in the hyperpneic phase between obstructive events (Vdrive). Without arousal, Vdrive is assumed to reflect chemical drive at the end of apnea. Arousal, if present, is assumed to contribute a fraction of Vdrive. A model with four variables (LG, time constant of chemical responses, circulatory delay, and the arousal contribution) is used to partition Vdrive into the components related to arousal (a constant predicted from the model; Varousal) and that related to the chemical drive (Vchem). The model is then used to estimate the time course of the chemical drive during the obstructed phase. From this LG, the chemical drive preceding arousal (arousal threshold), airway collapsibility, and pharyngeal muscle compensation are derived [8]. In our view, this model is based on untenable assumptions and does not identify the traits that are truly relevant to ventilatory instability in OSA: (1) Measurement of LG: (1a) A fundamental assumption in this model is that ventilation in the hyperpneic phase reflects respiratory drive (i.e. respiratory muscle output) as it does in participants without OSA. This is clearly not a valid assumption. Conversion of respiratory muscle output to ventilation is highly dependent on upper airway (UA) resistance. Resistance in the hyperpneic phase in patients with OSA cannot be assumed to be normal as it ranges from normal to very high [9], and it increases further beyond the first breath [9]. Thus, Vdrive and, by extension, LG and all “traits” derived from the model based on Vdrive is underestimated by an unknown amount. (1b) Another fundamental assumption is that the contribution of arousal to ventilation in the hyperpneic phase is the same (constant). This is again untenable. The arousal-mediated increase in ventilation is highly variable at different times in the same participant [10]. Furthermore, arousal intensity and duration vary considerably from time to time in the same participant [11], (1c) It is assumed that the ventilatory response to the obstructive event is represented by the respiratory drive in individual breaths within the hyperpneic phase. The ventilatory response relevant to calculating LG is not determined by respiratory drive in a single breath, or by the average respiratory drive in the hyperpneic phase, but by the excess ventilation generated by all breaths in the hyperpneic phase. It is this total excess ventilation that determines the extent to which post-event drive will be reduced and, by extension, event recurrence. The number of hyperpneic breaths varies considerably between events and between patients depending on lung-carotid circulation delay. Although pulmonary gas tensions improve simultaneously with the first open breath, the improvement is not sensed by the chemoreceptors for 1–3 breaths (circulation delay) during which the chemoreceptor continues to respond to asphyxic blood. Furthermore, the amplitude of the various breaths during the hyperpneic phase is influenced by the breath-by-breath changes in UA resistance [9]. Failure to include the cumulative excess ventilation in all breaths in the hyperpneic phase results in further underestimation of LG. (1d) LG is a ratio of the magnitude of the ventilatory response (ΔVERESPONSE) to the decrease in ventilation in the obstructed phase (ΔVE) [12]. In the authors’ approach [6, 7] ΔVE is based on the reduction in ventilation from average ventilation measured across events, including the ventilation in the obstructed breaths. This approach for calculating LG assumes that the apneic thresholds in patients with OSA are the same as the chemical apneic threshold that is relevant to central sleep apnea [12, 13]. This chemical apneic threshold is a lower eupneic drive [13]. This apneic threshold is not relevant to LG in OSA. In OSA, obstruction occurs above the chemical apneic threshold as evidenced by continued strong efforts (i.e. drive well above eupneic levels) during the obstructed phase [14]. What is relevant to LG in OSA is the chemical drive required to keep the airway open, which was called Effective Recruitment Threshold (TER) [2, 5]. In OSA patients TER ranged from eupneic level (drive present during stable breathing with an open airway) to >274% of a eupneic level, and was greater than the arousal threshold in 9 of 21 patients [5], indicating that the chemical drive required to keep the airway open in these patients is not consistent with sleep. This extremely relevant trait is completely ignored by the author’s approach. Given these considerations, it is not surprising that measured LG was <1.0 in nearly all patients [6]. If these values were accurate, why were these patients having recurrent events, during which LG is by definition >1.0? The authors do not address this contradiction. (2) Measurement of arousal threshold: As indicated in section 1b above, this measurement is affected by errors in the estimate of Vdrive. In addition, the arousal threshold can only be determined from experimentally induced arousals applied randomly throughout the PSG. No such arousals occur in routine sleep studies. Whether measured by delta power [15] or the odds ratio product [16], arousability changes continuously within and between respiratory cycles [15, 16]. The use of OSA-associated arousals to measure arousal threshold underestimates the arousal threshold in that such arousals will preferentially occur during periods of low threshold. Thus, a patient who spends 80% of the time in deep sleep, where there are few if any arousals, but has OSA with recurrent arousals in a fraction of the night when arousal threshold is low, will be assigned a low arousal threshold. In such patients, OSA may be attributed to a low arousal threshold (suggesting a potential benefit from sedatives) when the patient’s arousal threshold is quite high most of the time, and this is protecting the patient from having a much higher AHI. (3) Airway collapsibility: it is not possible to measure the passive collapsibility of the airway using the Terrill/Sands approach [7]. Measurement of passive collapsibility (passive PCRIT) requires measurement of inspiratory flow at least at two airway pressure points when dilator activity is absent or minimal [17]. Airway pressure is constant in diagnostic PSGs and there is no single point when dilator activity can be reliably considered absent or minimal. The standard approach [17] is based on prior observations indicating that dilator activity is minimal on CPAP [18, 19] and does not increase for 2–3 breaths when CPAP is dropped to atmospheric and obstruction results [17]. Moreover, PCRIT can be very difficult to measure and the exact methods for measuring PCRIT are unclear [20]. It should also be noted that PCRIT is not a true measure of upper airway anatomy. (4) Pharyngeal muscle compensation: This is the most problematic of the traits estimated by the Terrill/Sands approach [6, 7]. It is defined as the difference between Vactive and Vpassive. Vpassive is determined from actual ventilation measured when the estimated drive, using their model (but see the critique of this model, above), is equal to the eupneic drive. Justification given for this assumption is that the airway is passive at eupneic drive. In the references cited in support of this statement [8, 21] eupneic drive was defined as the ventilation on therapeutic CPAP, where the airway is patent throughout, and this is a correct statement [17–19]. However, in their application [7] eupneic drive is defined as the average ventilation of all breaths, obstructed and not obstructed. Clearly, this eupneic drive is not the same as the eupneic drive on therapeutic CPAP since the average includes breaths with high respiratory drive. Accordingly, Vpassive, so determined, greatly overestimates ventilation in the absence of dilator activity. Vactive is defined as the actual ventilation measured just before arousal [7]. It is not clear how this can be justified given the flow limitation that seriously constrains ventilation just before arousal. That these estimates are unrealistic is clearly shown in their results [7]. Thus, in 27 of the 31 patients, Vpassive was ≥70% eupneic ventilation [7]. If that were true, how did these 27 patients qualify for the diagnosis of OSA when hypopneas require a minimum reduction in the flow of 30%? Also, the so-called compensation (Vactive − Vpassive) was near zero in most patients [7], implying virtually no recruitment of UA dilators, which is clearly inconsistent with what is well established [1]. In some patients, compensation was even negative [7]. How can UA dilator muscle activity decrease below that at Vpassive when there was, by definition, no such activity at Vpassive? One might expect that such a potentially transformative proposal should be properly validated against gold-standard methods for measuring these traits. Instead, in the original description [6], validation was by showing that when LG is expected to decrease by O2 breathing or by acetazolamide the estimated LG did decrease [6]. Qualitative agreement is hardly convincing. They also performed Spearman correlation of the traits determined by the new method and those of an earlier model that used LONG (3-min) hypopneas induced by reducing CPAP [8]. This approach has its own problems one of which is that responses to long hypopneas are not appropriate for validation, since they incorporate large contributions from central chemoreceptors which have substantially different dynamics. Furthermore, correlation is not the appropriate way to evaluate the numerical agreement between two methods of measuring the same variable. When dealing with LG it is the actual value that is important. In the plot shown to validate the new method r was modest (0.69) [6] and there was much scatter. For example, LG of 0.3 by one method could be 0.8 by the other. These two values represent very different levels of instability. In a more recent paper, the authors measured diaphragm EMG and determined the conversion factor of EMG to VE under conditions of an open airway [7]. The intention was to show that when chemical drive during obstructive events is estimated from the EMG (using the wake conversion factor) it would match the time course estimated by their model. Such an agreement would have been convincing. However, they showed no quantitative results comparing diaphragm activity with model estimates of respiratory drive, and the selected tracings shown were not terribly reassuring [7]. For example, the actual and model-predicted VE differed substantially in some of the few illustrated breaths in the absence of arousal (e.g. first and fourth events in Figure 1, A) [7], yet were quite similar when arousal was present (e.g. first and third events in Figure 1, B, and first event in Figure 1, C) [7]. Furthermore, the increase in the model estimate of ventilatory drive did not properly match the actual increase in diaphragm EMG (e.g. doubling of the model estimate of drive for a sixfold increase in EMG; second event in Figure 1, C) [7]. Given these uncertainties, it is not surprising that despite 7 years of extensively using this model in OSA patients there has not been a single report in which traits determined using this method were used successfully to treat a patient at home. In addition, it should be noted that there are no proper reproducibility studies for any of the endophenotypes and, as discussed above, there are very limited validation data of the endophenotypes against gold standard methods. Phenotyping pharyngeal pathophysiology using polysomnography is based on a simple model of ventilatory control and as discussed above many of the model’s assumptions for all four traits are untenable. Moreover, independent of the flaws in the model assumptions and measurements, determining a single measure for airway collapsibility, LG, arousal threshold, and muscle responsiveness is also problematic since these values will vary throughout the night based on the stage of sleep, body/neck position, and other factors. Another practical limitation of PSG physiological endotyping is the need to use oronasal masks to measure the traits [6, 7]. A large number of patients with OSA exhibit mouth breathing. The reliance on the nasal cannula without a measurement of the oral ventilation will result in systematic errors in the endotype traits. It is not clear if this has been adequately addressed in the previous publications examining the physiologic traits. Oronasal thermistors can be used to detect oral breathing. However, their output is not proportional to flow making these devices inappropriate for determining ventilation. To move forward the assumptions from which the physiologic phenotypes are derived need to be much more solid and reproducibility studies must be performed at two different time points. Otherwise, at least based on the present data, these phenotypes should be considered a “house of cards.” The following was prepared in response to the opposing argument. We read with considerable interest the arguments by Sands and Edwards in favor of determining OSA phenotypes from clinical polysomnography (PSG) [22]. We totally agree that having a tool that can readily and simply identify the reasons why OSA is present, or why it is severe, in a particular patient (precision medicine), would be a most welcome and transformative development. What we disagree with is that analyzing data from routine PSGs can serve this purpose. To begin with, the current debate is not about extracting information from clinical PSGs that would suggest that a single endotype is high or not (last major paragraph in the PRO account) [6, 22]. This kind of information is not very informative in terms of patient management. For example, a high LG inferred from the presence of central or mixed apneas does not tell us how high it is or why the overshoot is due to a high controller gain, long circulatory delay, high arousal threshold that delays upper airway opening, or high drive required to open the airway (effective recruitment threshold; TER). Likewise, high collapsibility inferred from therapeutic CPAP level, or presence of apneas versus hypopneas, does not explain why breathing in such a patient is unstable when others with similar collapsibility are not [4]. It is this kind of detail that is needed to practice individualized patient management since these different mechanisms of high LG call for different therapeutic approaches. What the debate is about is whether the approach advocated by Sands and Edwards [6, 22], can provide the information needed to plan effective management. Sands and Edwards [22] nowhere acknowledge in their PRO manuscript that their proposed approach to measuring LG, arousal threshold, airway collapsibility, and muscle responsiveness violates some of the most fundamental tenets in respiratory physiology and engineering. These assumptions and their relevance to measuring endotypes by the Sands and Edwards approach were discussed in detail in our original CON manuscript, and briefly include: - Flow measured at one end of a tube is not only dependent on the pressure applied at the other end (thoracic pressure), but also on the resistance of the tube. Thus, without knowing the actual resistance of the airway, which cannot be assumed to be normal or even constant within the same event, flow measured at the face does not reflect respiratory drive. - LG is not determined by the amplitude of a single breath during the hyperpnea phase but by the total excess ventilation accrued during the hyperpnea phase, and hence by circulation delay and changes in resistance during that phase. It is the reduction of chemical drive during the unobstructed phase that is responsible for instability, and this reduction is a function of the excess ventilation during the entire unobstructed phase and not the size of just one breath. - Arousal intensity is not constant across all respiratory events. So, modeling its impact on ventilation as a constant across all events introduces an event-by-event variable error in the estimated arousal-independent increase in ventilation. This variable error can corrupt the model estimates of the various endotypes. - For estimating LG, the authors utilize a model developed to explain central apnea, which occurs when PCO2 decreases below the eupneic PCO2 level to the point of completely inhibiting respiratory muscle activity (i.e. diaphragm) resulting in apnea with an open airway. In OSA, apnea occurs at eupneic or above eupneic drive levels. Thus, the model used by Sands and Edwards is simply inappropriate for OSA as it does not consider the chemical drive required to keep the airway open [23]. - The arousal threshold is highly variable across the night and measuring it from the drive at the time of spontaneous (including OSA-induced) arousals, biases the measured threshold too low values and provides a false indication of the general arousability of the participant. - It is not possible to measure airway collapsibility from the PSG since this requires measuring flow when the pharyngeal muscles are completely relaxed, and this can never be ascertained in clinical PSGs [23]. - It is not possible to measure pharyngeal muscle compensation in clinical PSG since in the presence of flow limitation or obstruction in flow in the obstructed breaths may decrease or not in the face of considerable pharyngeal muscle compensation [23]. In to these measurement there are clear (e.g. mouth breathing with the measurement of the endophenotypes from routine polysomnography Moreover, independent of the flaws in the model assumptions and and determining a single measure for airway collapsibility, LG, arousal threshold, and muscle responsiveness is also problematic since these values will vary throughout the night based on the stage of sleep, body/neck position, and other factors. more is the absence of data showing validation gold standard and reproducibility two different time of these Without data showing validation and reproducibility it is to have confidence in the Sands and proposed approach has not been validated against the gold standard methods of measuring these this should have been performed many years In Sands and Edwards suggest that agreement with values generated by them using an approach is validation Sands and Edwards studies that showed that one endotype or determined by the authors’ was different in the expected between two with different or clinical that their data are these are not validation studies against a gold have been recent studies on the stability of the endophenotypes and these studies do not show strong reproducibility of the traits. The first an performed two PSG studies in participants at in to the reproducibility of the The from the were that LG and arousal thresholds may as stable of physiological but collapsibility and compensation do not to stable physiological These data at of the endophenotypes are and problematic are recent data published by and of The of and in The authors using of traits derived from and periods and was using the in at points years for traits correlation ranged from to these data only provide information on the stability of the traits throughout the not information on whether a value during a given night is Moreover, examining the on and even periods is of in an of with confidence the endotypes are consistent within a night than the AHI. The points years was also in the with results only modest The were collapsibility muscle compensation LG and arousal threshold with only LG the with this of agreement suggest that values of all endophenotypes show large between the points. For example, despite similar average values at and data show that in LG may from to this of the values and is nearly of the Thus, own data do not provide in support of the reproducibility of the Moreover, Sands and Edwards [22] PRO manuscript is with of studies that have no relevance to the of the model or its use in One is a that a higher LG in The average for was for it was indicating a in the LG in the two the were different at simply about participants of were used in the How does this information validate the use of the model in a single Another approach to the between results in two is to a the at the endotype level below or above which there is no and that this level who will or will not respond to an How can that if any single whether or can have any value within the possible of the Another is a by that agreement between results generated by the used by the authors and results of a in a different which simply that was to the authors’ results In cited the authors agreement between results studies and PSG even LG could by to between the two values in the same individual In of the of validation and reproducibility in the of LG, arousal threshold, airway collapsibility, and muscle responsiveness on data from routine polysomnography in these this of should be considered at a in and at a “house of cards.” we the question we in our CON manuscript if the model results are valid why have there not been any of of OSA patients using by the proposed No new data were generated or in support of this This is a debate between the of this and the of the opposing argument.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.