A simplified method for determining phenotypic traits in patients with obstructive sleep apnea

Andrew Wellman, Bradley A. Edwards, Scott A. Sands, Robert L. Owens, Shamim Nemati, James Butler, Chris L. Passaglia, Andrew C. Jackson, Atul Malhotra, David P. White


We previously published a method for measuring several physiological traits causing obstructive sleep apnea (OSA). The method, however, had a relatively low success rate (76%) and required mathematical modeling, potentially limiting its application. This paper presents a substantial revision of that technique. To make the measurements, continuous positive airway pressure (CPAP) was manipulated during sleep to quantify 1) eupneic ventilatory demand, 2) the level of ventilation at which arousals begin to occur, 3) ventilation off CPAP (nasal pressure = 0 cmH2O) when the pharyngeal muscles are activated during sleep, and 4) ventilation off CPAP when the pharyngeal muscles are relatively passive. These traits could be determined in all 13 participants (100% success rate). There was substantial intersubject variability in the reduction in ventilation that individuals could tolerate before having arousals (difference between ventilations #1 and #2 ranged from 0.7 to 2.9 liters/min) and in the amount of ventilatory compensation that individuals could generate (difference between ventilations #3 and #4 ranged from −0.5 to 5.5 liters/min). Importantly, the measurements accurately reflected clinical metrics; the difference between ventilations #2 and #3, a measure of the gap that must be overcome to achieve stable breathing during sleep, correlated with the apnea-hypopnea index (r = 0.9, P < 0.001). An additional procedure was added to the technique to measure loop gain (sensitivity of the ventilatory control system), which allowed arousal threshold and upper airway gain (response of the upper airway to increasing ventilatory drive) to be quantified as well. Of note, the traits were generally repeatable when measured on a second night in 5 individuals. This technique is a relatively simple way of defining mechanisms underlying OSA and could potentially be used in a clinical setting to individualize therapy.

  • pathophysiology of sleep apnea
  • loop gain
  • pharyngeal critical closing pressure
  • upper airway
  • arousal threshold

recent evidence suggests that obstructive sleep apnea (OSA) is a multifactorial disorder. Contributing factors include a small or collapsible pharyngeal airway (9, 12, 23), a high loop gain (large ventilatory response to a ventilatory disturbance) (2, 11, 21, 25, 26, 31), poor pharyngeal muscle responsiveness during sleep (5, 1416, 20, 27, 29), and a low respiratory arousal threshold (13, 28). The relative contribution of these traits varies substantially between individuals (24, 30).

Despite the multifactorial nature of OSA, common therapies [continuous positive airway pressure (CPAP), upper airway surgery, dental appliances] are directed at only one trait: the abnormal airway anatomy. Moreover, the most effective treatment, CPAP, has an acceptance rate of only ∼50% (8, 17). We believe that if the traits causing OSA could be measured in a clinically feasible way, then alternative therapies might be possible in some patients. For example, drugs for lowering loop gain (7) or raising the arousal threshold (6) might be useful in patients for whom these are a problem.

We recently published a procedure for measuring the four traits described above (24). This study is a follow-up to that paper in which the technique has been revised to overcome a major problem. Figure 1 gives a brief explanation of the previous technique. The main problem with that approach was that some individuals fully compensated during the 3-min CPAP drops, and thus it was difficult to produce a disturbance (denominator in the loop gain ratio) without arousing the individual. Moreover, when loop gain could not be determined, the other traits could not be determined either, and thus the subject could not be phenotyped. When tested in 79 people, 19 could not be adequately phenotyped with the previous method.

Fig. 1.

Schematic illustration of the original method for phenotyping OSA. The primary maneuver is a 3-min CPAP drop. On optimum CPAP (10 cmH2O in this example), the airway is completely patent and ventilation (a) is assumed to meet the eupneic ventilatory demand during sleep. When CPAP is dropped, the upper airway narrows and decreases ventilation from 6 to 2.7 liters/min (b). The decreased ventilation leads to an increase in carbon dioxide and hence ventilatory drive that stimulates the diaphragm and the pharyngeal muscles. In many individuals, such stimulation allows ventilation to recover slightly (c) but not completely (although in some patients there is complete recovery as described in the text) and a new steady state below eupnea is reached (4.6 liters/min in this example). When CPAP is returned to the optimum pressure, the airway opens and the increased ventilatory drive manifests as a ventilatory overshoot to 10.2 liters/min (d). Eventually, ventilation and ventilatory drive return to eupnea (e) because the airway is now patent and ventilation can once again match the ventilatory demand. Loop gain is calculated by dividing the response (the ventilatory overshoot of 4.2 liters/min) by the disturbance (the steady state reduction in ventilation of 1.4 liters/min). Loop gain = 4.2 liters/min ÷ −1.4 liters/min = −3.

The new method described in this paper overcomes this difficulty primarily by shortening the time required to measure each trait, including loop gain. As will be described below, this shorter measurement time allows CPAP to be decreased to a lower level for longer periods of time, increasing the likelihood that a disturbance in ventilation can be produced without arousing the subject. Moreover, the new approach provides a framework for quantifying the other physiologic traits when loop gain is not, or cannot, be determined. Finally, other important changes have been made that simplify the analysis and interpretation of the data.



CPAP-treated OSA patients (>5 h/night for >2 mo) were recruited from our clinical sleep laboratory. OSA was defined as an apnea-hypopnea index (AHI) >10 events/h during supine, non-rapid eye movement (NREM) sleep. Normal subjects were recruited from the community and had a supine NREM AHI <10 events/h. Subjects were on no medications known to affect respiration or pharyngeal muscle control. Exclusion criteria included concurrent sleep disorder, renal insufficiency, neuromuscular disease, uncontrolled diabetes mellitus, central sleep apnea, heart failure, uncontrolled hypertension, or thyroid disorder. The age range was 21–65 yr. The study was approved by the Institutional Review Board at Brigham and Women's Hospital. All subjects gave informed, written consent before participating.


Standard polysomnography.

Standard polysomnography (PSG) was performed in the clinical sleep laboratory with at least 4 h of supine sleep recorded. Electroencephalography (EEG), chin electromyography (EMG), and electrooculography (EOG) were recorded for sleep staging. Airflow was measured using nasal pressure and a thermistor. Piezo-electric bands around the chest and abdomen monitored respiratory movements. Electrocardiography, tibialis EMG, and arterial oxygen saturation were also measured. Signals were collected using the Alice 5 data acquisition system (Philips Respironics, Murrysville, PA). Apneas and hypopneas were scored using standard American Academy of Sleep Medicine guidelines (1), and the AHI reported is the value during supine, NREM sleep.

Physiological PSG.

On a subsequent night within ∼1 wk, the physiologic traits were measured in the research laboratory. Subjects breathed through a sealed nasal mask attached to a modified CPAP machine (Philips Respironics) capable of delivering pressures between +20 and −20 cmH2O. EEG, chin EMG, and EOG were measured for sleep staging. Airflow was measured with a pneumotachometer (Hans-Rudolph, Kansas City, MO) and a pressure transducer (Validyne, Northridge, CA) attached to the sealed nasal mask. Mask pressure was monitored with a pressure transducer (Validyne) referenced to atmosphere. Arterial oxygen saturation at the finger (BCI, Waukesha, WI) and exhaled carbon dioxide in the mask (Vacumed, Ventura, CA) were also measured. Signals were sampled at 125 Hz and recorded using Nihon Khoden (Tokyo, Japan) and Spike 2 software (Cambridge, UK).

Measurement of the Physiological Traits

Subjects slept in the supine position. Once asleep, CPAP was set to the patient's prescribed pressure or, for the normal subjects, 3–4 cmH2O. The pressure was increased as needed to eliminate residual snoring, hypopneas, or flattened inspiratory flow. CPAP was then manipulated as follows to measure the physiological traits. First, the mask pressure was rapidly dialed down to 0 cmH2O for 5 breaths (see passive drop in Fig. 2). Following a return to the optimum level, the CPAP was then decreased slowly (using the algorithm described in Fig. 3) to determine the level of ventilation at which arousals begin to occur (4). The CPAP associated with this level of ventilation was termed the minimum tolerable CPAP, or CPAPmin (in some subjects this was a negative pressure). The CPAPmin is similar to the cycling threshold described by Chin et al. (4). CPAP drops to 0 cmH2O (active drop in Fig. 2) and dials up to optimum pressure (dial-up in Fig. 2) were performed during periods of relatively arousal-free breathing on CPAPmin. If awakening occurred on suboptimum CPAP, it was returned to the optimum level until sleep resumed. Several reduction sequences like the one shown in Fig. 2 were performed during the night.

Fig. 2.

Schematic drawing showing how CPAP is manipulated to measure the traits. Starting from an optimum pressure, CPAP is rapidly dialed down to zero for 5 breaths (Passive drop). It is then slowly decreased by 1 cmH2O per minute until the minimum tolerable CPAP (CPAPmin) is achieved. Both active CPAP drops and dial-ups are performed from CPAPmin. Calculation of the different traits from these maneuvers is described in the text.

Fig. 3.

Algorithm for achieving the minimum tolerable CPAP. CPAP is decreased by 1 cmH2O every minute until a respiratory effort–related arousal occurs (the algorithm does not stop if a spontaneous arousal occurs, see text for definition of a spontaneous arousal). If after the first arousal another respiratory arousal occurs within 1 min, the pressure is increased by 0.5 cmH2O. If not, then a CPAP drop or dial-up is performed. Once two measurements apiece are made at a particular CPAPmin, then it is decreased by 0.5 cmH2O and the loop is repeated. The goal is to decrease CPAP to a level that just begins to cause arousals, but the arousals cannot be so frequent that a quasi–steady state cannot be achieved between the arousals.

The phenotypic traits were quantified as four ventilation parameters that were measured from these manipulations in CPAP. The definition of each and their significance in the pathogenesis of OSA are described below. More detailed instructions for processing the signals, along with computer programs for automating the analysis, are provided in the online supplement.

Eupneic ventilation (Veupnea).

The first ventilation parameter is Veupnea, which is the subject's asleep ventilatory requirement based on metabolic rate and dead space ventilation. It was determined by averaging several minutes of ventilation while the subject was asleep and during stable breathing on optimum CPAP.

Ventilation that leads to arousals (Varousal).

When ventilation decreases during the slow CPAP reduction, carbon dioxide and ventilatory drive increase and can cause respiratory effort–related arousals. The ventilation leading to such arousals is called Varousal. Varousal was measured as follows: 1) When an arousal occurred on any pressure below the optimum CPAP, the 5 breaths prior to the arousal were inspected for flow limitation, defined by visual inspection as a peak-plateau inspiratory flow shape with prolonged inspiratory time (22). 2) If there was no clear flow limitation, or if the ventilation was greater than Veupnea, then the arousal was considered spontaneous and the data were not used to calculate Varousal. 3) If all 5 breaths were clearly flow-limited and the mean ventilation of the 5 breaths was less than Veupnea, then the mean inspiratory flow rates were next inspected to determine whether the smallest breath differed from the largest breath by >50%. If so, these data were not used for Varousal to ensure that a relatively stable reduction in ventilation existed prior to arousal. If not, then the average ventilation of the 5 breaths preceding the arousal was used for Varousal. 4) Other reasons for excluding data from the Varousal calculation included: a) if an arousal occurred within the previous 60 s, thus Varousal was not determined from data obtained during cyclic hypopneas; b) if the arousal occurred as a result of an active CPAP drop; and c) if an apnea (zero flow) led to the arousal. Varousal measurements were made for each arousal meeting criteria.

Important physiological information can be obtained from Veupnea and Varousal. The difference between these values is the amount of reduction in ventilation that an individual will tolerate before arousing from sleep. This difference is a function of the arousal threshold. Arousals from sleep occur when ventilatory drive (or respiratory effort) reaches a certain threshold, termed the arousal threshold (3, 10). Individuals with a low arousal threshold (arousal occurs at a low level of ventilatory drive) should have a relatively high Varousal and thus a small difference between Veupnea and Varousal. Those with a high arousal threshold, on the other hand, should tolerate a low Varousal and have a large difference between Veupnea and Varousal.

The difference between Veupnea and Varousal is also a function of the respiratory control system loop gain. Loop gain is the amount of increase in ventilatory drive that occurs for a given reduction in ventilation (the loop gain referred to here is the steady state loop gain; i.e., the increase in ventilatory drive to a stable, as opposed to a transient, reduction in ventilation). A high loop gain means that there is a large increase in ventilatory drive for a small reduction in ventilation, which would tend to raise Varousal and produce a small difference between Veupnea and Varousal. A low loop gain, on the other hand, would tend to lower Varousal and permit a large difference between Veupnea and Varousal. To determine how much arousal threshold and loop gain contribute to the Veupnea minus Varousal difference, an additional maneuver can be added to the procedure to determine loop gain. This is described below.

Ventilation at a nasal pressure of 0 cmH2O when the pharyngeal muscles are passive (passive V0).

On optimum CPAP and at the eupneic level of ventilatory drive, the pharyngeal muscles are hypotonic or relatively passive. A rapid reduction in pressure to 0 cmH2O will show how much ventilation an individual can achieve through the passive airway. Thus, this parameter measures the anatomical predisposition to upper airway collapse. A passive V0 near Veupnea indicates a low predisposition to collapse (good anatomy), whereas a passive V0 at or near 0 liters/min indicates a high propensity to collapse (poor anatomy).

Passive V0 was measured by performing multiple CPAP drops from the optimum pressure (see Passive drop in Fig. 2). The pressure was rapidly dropped during expiration and returned to the optimum pressure during expiration of the fifth breath of the drop. Because breaths 1 and 2 of a drop are often confounded by lung volume changes (18, 19), breaths 3 and 4 were used for analysis. If zero flow was observed on the first breath, then mask pressure was held at 0 cmH2O for a total of 10 s before returning to the optimum pressure (to avoid prolonged apneas and to minimize awakenings).

Ventilation at a nasal pressure of 0 cmH2O when the pharyngeal muscles are active (active V0).

On CPAPmin, when ventilatory drive is at or near the arousal threshold, the pharyngeal muscles may become active and stiffen the airway, thereby permitting more airflow at 0 cmH2O than during the passive condition. Active V0 was measured by dropping CPAP to 0 cmH2O while subjects were breathing on CPAPmin (see Active drop in Fig. 2). Again, CPAPmin was achieved using the algorithm in Fig. 3. At CPAPmin, the muscles are stimulated as much as possible during sleep (any greater stimulation would lead to arousal). It is inevitable that occasional arousals will occur at CPAPmin due to fluctuations in ventilation and arousal threshold. Nevertheless, it is generally possible to find a level of CPAP at which arousals are infrequent (approximately a few minutes apart), with intervening periods of relatively stable breathing. It was during these periods of relative stability that active drops were performed. It was required that no arousals occur for 60 s prior to the drops (i.e., they were not performed in the midst of cyclic hypopneas). Only drops from the lowest CPAP level were used for determining active V0. If ≤3 drops were performed at the lowest level, then data from the next lowest level were also included. A raw data example of the active and passive V0 measurements is shown in Fig. 4.

Fig. 4.

Measurement of passive and active V0 from CPAP drops in one subject. Left: a CPAP drop from the optimum pressure. The ventilation decreases to 0 liters/min due to airway closure. When the drop is repeated from the minimum tolerable CPAP (right), the pharyngeal muscles are more active and the ventilation drops to 3.6 liters/min.

The difference between active and passive V0 is a measure of upper airway responsiveness, or the ability to stiffen/dilate the pharyngeal airway in response to an increase in ventilatory drive. It is also a function of arousal threshold. A high arousal threshold will permit more ventilatory drive to accumulate without an arousal, which can be used to further stiffen the airway and increase active V0.

The four ventilation measurements described above can be plotted as shown in Fig. 5, which shows typical values for a subject with OSA. Such a plot shows the relationship between the parameters and illustrates their physiological importance. As stated above, the difference between Veupnea and Varousal (a in Fig. 5) is the amount of reduction in ventilation that an individual will tolerate without waking up, which is a function of loop gain and arousal threshold. The difference between active and passive V0 (c in Fig. 5) is a measure of upper airway responsiveness and arousal threshold. Finally, the difference between Varousal and active V0 (b in Fig. 5) is the gap that must be overcome to achieve stable breathing during sleep. If active V0 is less than Varousal, then stable breathing will not occur because ventilation cannot increase above the level needed to prevent arousal; the patient will have OSA. If active V0 is greater than Varousal, then stable breathing (no OSA) should occur because ventilation can increase above the minimum tolerable level.

Fig. 5.

Graphical display for interpreting of the four ventilation parameters. Values for a hypothetical subject are shown. In A, the gaps between the ventilation parameters, marked a-c, have physiological importance as described in the text. Stable breathing could theoretically be achieved by decreasing Varousal below active V0 (B) or raising active V0 above Varousal (C).

Additional maneuver to determine loop gain.

This is considered an additional maneuver because we wanted to develop a technique that did not require loop gain measurement to interpret the results, as the previous method did. Loop gain can be measured as follows. During flow-limited breathing at CPAPmin, when drops for active V0 would typically be performed, CPAP is dialed up to the optimum pressure for 3 breaths (see Dial-up in Fig. 2). During flow limited breathing at CPAPmin, breathing is generally below eupnea, which leads to an increase in Pco2 and the drive to breathe. The CPAP dial up reopens the airway and reveals the full extent of the patient's ventilatory drive in response to the reduced ventilation on CPAPmin. This maneuver (CPAP dial up) allows the ventilatory drive response to a reduction in ventilation (loop gain) to be measured. An example is shown in Fig. 6. Loop gain is calculated as a ratio: the numerator is the ventilation on the CPAP dial up minus Veupnea; the denominator is Veupnea minus the ventilation at CPAPmin immediately before the dial up.

Fig. 6.

Measurement of loop gain from a CPAP dial-up in one subject. The eupneic minute ventilation (Veupnea), obtained from breathing on optimum CPAP (left), is 6.8 liters/min. At the minimum tolerable CPAP (right), the ventilation was 5.3 liters/min. When CPAP was dialed up to the optimum pressure for 3 breaths, ventilation increased to 16.6 liters/min. Therefore, the loop gain is (16.6 − 6.8)/(6.8 − 5.3) = −6.5.

Once the loop gain is known, it can be graphically represented with the four ventilation measurements to further delineate the phenotypic traits. In Fig. 7A, the loop gain (determined in Fig. 6) is plotted on a graph of ventilation vs. ventilatory drive. The reason for plotting loop gain on these axes is to show its relationship to the other traits, which will be added as described below. Ventilatory drive is chosen as the x-axis because it is measureable from the ventilatory overshoot and it is the relevant output variable for the ventilatory control feedback loop. First, Veupnea is placed along the line of identity between ventilation and ventilatory drive because it indicates that the patient's ventilatory demand is being fully met when the airway is open with CPAP. Loop gain is then plotted by starting at Veupnea, and drawing the reduction in ventilation of 1.5 liters/min (downward arrow), which produces a 9.8 liter/min increase in ventilatory drive (horizontal arrow). The line connecting the arrows is the reciprocal of loop gain (loop gain = change in ventilatory drive ÷ change in ventilation). Next, arousal threshold can be determined as shown in Fig. 7B. A horizontal line is drawn through Varousal, which in this example is 5.1 liters/min. The intersection of that line with the loop gain line is the arousal threshold, which is represented by a vertical line. This means that an arousal will occur when ventilatory drive reaches 18 liters/min. Lastly, the upper airway response is determined by first drawing a horizontal line through active V0 and noting where it intersects the arousal threshold. The passive V0 is then connected with this intersection. The slope of this line is called the upper airway gain (upper airway gain = change in ventilation ÷ change in ventilatory drive). The upper airway gain in this example is 0.4. In many individuals, the upper airway gain is probably nonlinear, exhibiting a “dog leg” shape as well as hysteresi (e.g., there may be no increase in ventilation until drive reaches a threshold, and, in some cases, ventilation may remain elevated even if drive later drops below the threshold). However, it is represented to the first approximation here as a straight line. The line is dashed to indicate this simplification.

Fig. 7.

OSA model diagram for an example subject (see text for detailed discussions of the plots). A: plot of loop gain. B: arousal threshold can be obtained from the loop gain and Varousal. C: the upper airway gain can then be drawn using the passive and active V0. D: interpretation of the model. E: effect of a small increase in passive V0. F: effect of an increase in both passive V0 and arousal threshold. LG, loop gain; ArThr, arousal threshold; UAG, upper airway gain.

Figure 7D is a two-dimensional version of Fig. 4 and indicates that on optimum CPAP, ventilation is 6.8 liters/min; off CPAP, but still at the eupneic ventilatory drive, the airway shuts and ventilation drops to 0 liters/min. As ventilatory drive increases, ventilation increases approximately along the upper airway gain line (see dotted line). When ventilatory drive reaches the arousal threshold, an arousal occurs and the airway opens. Ventilation increases until the patient falls back to sleep and the airway obstructs and the cycle repeats as indicated by the dotted line (the patient has OSA). Fig. 7, E and F show how small adjustments in the traits might affect breathing. In Fig. 7E, passive V0 is increased by 1 liter/min to simulate a slight improvement in the pharyngeal anatomy as might occur with an upper airway surgical procedure or a dental appliance. OSA persists according to the model. If this improved V0 is combined with an increase in arousal threshold, however, OSA could potentially be eliminated because now the upper airway gain line intersects the loop gain line prior to the arousal threshold. Therefore, stable breathing can be achieved without arousal and there is no OSA. Ventilation and ventilatory drive stop when they reach the loop gain because this is the simultaneous solution to the equations for both lines. We have labeled this intersection the steady state point. Note that when the steady state point is to the left of the arousal threshold, then active V0 is above Varousal (i.e., the subject is able to achieve a stable level of ventilation above that which arousal occurs).


Subject characteristics are shown in Table 1; polysomnography parameters are shown in Table 2. Eight subjects were women, 5 were men (median age, 46 years; median BMI, 33 kg/m2). Subjects exhibited a broad range of apnea severity (2–75 events/h) and optimum CPAP (3–17 cmH2O). Subjects 1–8 were studied once; subjects 9–13 were studied twice for repeatability.

View this table:
Table 1.

Subject characteristics

View this table:
Table 2.

Polysomnography parameters

In total, 133 dial-ups, 162 CPAP drops, and 185 Varousal measurements were made. Phenotypic traits were obtained in all 13 participants, although in one individual (subject 5) CPAP dial-ups were not attempted because this was the first subject and the dial-up procedure had not been conceived. As a result, loop gain was not defined in this subject. Loop gain could be measured in all subjects in whom it was attempted. This is contrast to the previous method in which loop gain could be determined in only 60 out of 79 individuals. Thus although the new technique was designed to work when loop gain could not be measured, we in fact had better success measuring loop gain with the new method. The reason for this is addressed in the discussion.

Table 3 shows the median values and ranges for the traits (the differences between the ventilation parameters, rather than the absolute values, are presented because these are the most relevant to phenotyping). Subjects tolerated ∼1.4 liters/min reduction in ventilation during sleep before having arousals (Veupnea − Varousal), although there was a wide range of variability. The median compensatory ventilation (active V0 − passive V0) was 2 liters/min, but again, the median exhibited a broad range of variability, from no response to almost complete airway opening in some individuals. The large intersubject variability is also evident in the traits when expressed as loop gain, arousal threshold, and upper airway gain. Note that arousal threshold varies fourfold, and the loop gain and upper airway gain vary by an order of magnitude.

View this table:
Table 3.

Medians and ranges for phenotypic traits

Figure 8 displays the model results for subjects 1–8 in order of increasing AHI. Subject 1 has no OSA because of good anatomy (passive V0 is just below Veupnea), a good upper airway gain (0.5), and a normal loop gain (−3.6). The low arousal threshold (7.4 liters/min), which is a risk factor for OSA, is inconsequential in this subject. Similarly, in subject 2, the negatively sloped upper airway gain (airway sucked progressively closed with each successive breath as respiratory drive increases) is inconsequential because the anatomy is excellent (passive V0 equals Veupnea), and the low loop gain (−2.5) and high arousal threshold (13.2 liters/min) allow the patient to tolerate a 50% reduction in ventilation. The most remarkable feature in subjects 3 and 4 is the minimal OSA despite bad anatomy (passive V0 of 0.6 and 0.3 liters/min, respectively). In these individuals, activation of the pharyngeal muscles is highly effective (active V0 is 3.8 liters/min above passive V0 in both subjects, and thus the upper airway gain is very positive). Subject 5 was able to tolerate a significant reduction in ventilation (2.5 liters/min below Veupnea). It is not clear whether this was due to a low loop gain or a high arousal threshold, because loop gain was not determined. Nevertheless, this subject has good compensatory mechanisms and, in the fully compensated state during sleep, is close to achieving stability (the gap between Varousal and active V0 is 1.7 liters/min). Subject 6 had an excellent upper airway response but tolerated only a 0.7 liter/min reduction in ventilation due to a high loop gain (−6.7). These factors, along with the poor anatomy, are the likely cause of severe OSA. Lastly, subjects 7 and 8 have the unfavorable combination of both poor anatomy and poor airway response, and therefore, OSA would likely occur regardless of the other traits.

Fig. 8.

OSA model diagrams for subjects 1–8. See text for description. LG, loop gain; ArThr, arousal threshold; UAG, upper airway gain.

Figure 9 shows a side-by-side comparison of the traits determined on separate nights in subjects 9–13. For unclear reasons, subject 9 had a ventilatory demand of 6.3 liter/min on night 1 and 5 liters/min on night 2. This trait is probably the least likely to be affected by measurement noise, and thus the difference probably represents true biological variability. Nevertheless, the other traits appear similar on both nights: loop gain and arousal threshold are low, upper airway gain is slightly positive, and passive V0 is just below or the same as Veupnea. In subject 10, the differences between Veupnea and Varousal and active and passive V0 are similar on both nights, and the loop gain is low on both nights. The other traits are slightly different, but the general physiology on both nights is consistent with mild sleep apnea. Subject 11's traits were quite similar on both nights, with the upper airway gain being slightly more positive on the second night. Subject 12, like subject 9, had a significantly different Veupnea on the different nights. However, the other traits are comparable. Lastly, subject 13's traits were similar as well between the nights.

Fig. 9.

OSA model diagrams in subjects 9–13 showing repeatability from night 1 (left) to night 2 (right). LG, loop gain; ArThr, arousal threshold; UAG, upper airway gain.

A brief scan of Figs. 8 and 9 shows that the model was also fairly accurate; the difference between Varousal and active V0, a measure of the gap that needs to be overcome to achieve stable breathing, correlates closely with AHI (Spearman correlation coefficient = 0.9, P < 0.001; see Fig. 10). Thus, individuals without OSA (subjects 1, 2, and 9) have an active V0 well above Varousal. Those with mild OSA (subjects 3 and 4) had an active V0 slightly below or above Varousal. And in general, as OSA became more severe, the gap between active V0 and Varousal became larger. For completeness, correlation analyses between the other traits and AHI are provided in Table 4. However, the current study is a small feasibility study and thus underpowered to adequately test some of these associations.

Fig. 10.

Correlation between Varousal − active V0 and supine NREM AHI.

View this table:
Table 4.

Correlation between the traits and AHI

Lastly, measurements using both the original and revised techniques were performed in 7 individuals for comparison. Figure 11 shows the results, with the values being plotted along the line of identity. The loop gain values (Fig. 11A) show good agreement, which makes sense as these parameters are determined in a very similar manner between the two methods. The arousal threshold measurements (Fig. 11B) also agree. However, the upper airway gain (UAG) and passive V0 are surprisingly different. One possible reason for this discrepancy is that these parameters were determined quite differently in each method. The old method measured them by dropping CPAP to subtherapeutic (but supraatmospheric) levels, whereas the new method actually dropped CPAP to atmospheric levels. It is possible that the airway compensates differently at supraatmospheric compared with atmospheric pressure. Moreover, higher levels of ventilatory drive were likely achieved in the new method, which could affect how the upper airway gain is quantified. Of note, the passive V0 was systematically lower (worse passive anatomy) in the new technique. This discrepancy could potentially be due to differences in lung volume: the new procedure measures passive and active V0 at functional residual capacity (FRC; atmospheric nasal pressure), whereas the old procedure measured ventilation above FRC (supraatmospheric pressure) and extrapolated backward to determine ventilation at zero pressure.

Fig. 11.

Comparison between the old and new techniques. A: loop gain magnitude (dimensionless, the absolute values are plotted); B: arousal threshold (values are in liters/min above Veupnea); C: upper airway gain (UAG, a dimensionless variable); and D: passive V0 (liters/min).


This paper describes a significant revision of a previously presented approach for defining phenotypic traits in patients with OSA. The following revisions were made. First, rather than dropping CPAP to a suboptimum level for 3-min intervals, subjects were held at a low level of CPAP for several minutes at a time. This was the most significant change and was made possible by shortening the time needed to make individual measurements of the traits. Second, V0 was estimated by dropping CPAP to zero for 5 breaths. In the previous technique, V0 was determined from the y-intercept of a linear fit to ventilation vs. pressure data. Third, the traits were directly measured from the flow (or ventilation) signal, as opposed to estimating them by fitting data to a dynamic mathematical model.

The major findings are as follows: 1) the new technique is feasible and has a higher success rate than the previous technique; 2) the traits are generally repeatable but in some cases exhibit physiologic variability between nights; 3) the model diagram agrees with the AHI measurement (e.g., patients with mild OSA are close to achieving stable breathing in the model diagram, whereas those with severe OSA are not); and 4) the data are consistent with recent evidence that sleep apnea is a multifactorial disorder.

We believe the measurements are easier to obtain with the new technique, particularly in regards to loop gain, for the following reason. Typically at the CPAPmin (often about 4–5 cmH2O below the optimum level), inspiratory flow limitation develops and breathing remains relatively stable, albeit with intermittent arousals. The frequency of these arousals often varies; sometimes they are relatively far apart (e.g., every 4–5 min) while at other times they are close together (e.g., every 1–2 min). This is likely due to slight variations in arousal threshold during the course of the night. By leaving CPAP at or near the minimum tolerable level for prolonged intervals, we can wait for these periods of stable breathing to make measurements. Essentially, the shorter measurement time (3–4 breaths as opposed to 3–4 min) allows the traits to be measured between the arousals. It is worth mentioning that with any procedure that increases ventilatory drive (e.g., CO2 rebreathing or CPAP drops), the possibility exists of producing arousals, which can prevent measurements from being made. It is nevertheless necessary to increase drive as much as possible during sleep to effectively measure loop gain and upper airway compensatory mechanisms. Therefore, the measurements are inherently tricky. We believe that the revised approach maximizes the potential for making measurements by both leaving the pressure at a suboptimum level for longer periods of time and shortening the time needed to make each measurement.

Another key advantage of the revised method is that the traits are directly measured, as opposed to fitting data to a model. In the original technique, loop gain was estimated by fitting segments of ventilation data to a dynamic mathematical model. The analysis, therefore, required knowledge about mathematical modeling, potentially making this approach less accessible to some clinicians and physiologists. A principle aim of this research is to develop tools for measuring physiology at the bedside. We believe the revisions made are a significant step in this direction. By the same token, passive and active V0 are measured by directly dropping mask pressure to 0 cmH2O, rather than fitting a line to ventilation vs. pressure data and extrapolating to zero pressure. It is generally preferable to estimate parameters by directly measuring them in the region of interest, rather than making measurements outside the region of interest and extrapolating. We believe these improvements give the measurements more face validity, and they also provide clinicians with a more intuitive framework for analyzing and interpreting the results.

A major finding in our data is the wide range of variability in the traits. This has two important implications. First, for studies comparing traits between groups (e.g., those with and without OSA), which are often small physiological studies due to the complexity of the measurements, significant differences may be difficult to demonstrate. Second, it means that a trait need not be considered abnormal to contribute to OSA. Thus, a loop gain or arousal threshold that is within the normal range of variability could, in principle, play an important pathogenic role in some individuals.

The model diagrams (Figs. 8 and 9) also emphasize the multifactorial nature of OSA. Note that although all subjects with OSA (subjects 3–8 and 10–13) had a poor airway, the AHI varied substantially, and some individuals with a poor anatomy (subjects 3 and 4) had little or no OSA. Thus, pharyngeal collapsibility is only one of several components contributing to OSA. Furthermore, although we originally envisioned that individuals would likely separate into characteristic phenotype groups (e.g., those with OSA due primarily to anatomic problems, those with primarily a loop gain problem or arousal threshold problem, etc.), these results suggest that numerous possible combinations of traits can cause or prevent OSA. Indeed, it is difficult in many of the individuals to identify a single mechanism. Therefore, not only can OSA occur for different reasons, but several reasons may be at play in any one individual. For this reason, studies should focus on integrating the physiologic measurements rather than measuring a single trait.


Despite these improvements, there are still several limitations with this technique. First, CPAPmin cannot be set between approximately +1 and −2 cmH2O for an extended period. At these pressures, the total flow rate in the tubing is not high enough to flush CO2 from the mask, and consequently there is CO2 rebreathing. This is generally not a problem in patients with OSA in whom the minimum tolerable pressure is typically greater than +1 cmH2O. For some individuals without OSA, however, specialized breathing circuits will need to be developed to address this problem. Second, this method cannot distinguish differences in airway collapsibility between individuals with a highly collapsible airway. The airway is only tested at 0 cmH2O for the passive and active V0, and so the collapsibility at other pressures cannot be determined. Therefore, if the airway is closed at 0 cmH2O, it is not known whether it will also be closed at +1 or +2 cmH2O. If this information is needed, however, the procedure could be modified to drop CPAP to levels above zero. Third, the technique depends on achieving a minimum tolerable CPAP, which is somewhat subjective and varies throughout the night. We therefore tried to standardize the CPAP reduction as much as possible (see Fig. 3), although as more patients are studied, this procedure may require further refinement.

Finally, this method requires validation, although for some of the traits (loop gain and upper airway gain) it would be difficult because there is no gold standard for comparison. As stated, we do believe this technique has the advantage of face validity. For example, the ventilation that causes arousals (Varousal) is the actual ventilation before arousals, the ventilation at zero pressure (passive and active V0) is obtained by dropping the pressure to zero, etc. The traits can also be measured by hand; there are no internal variables or black box algorithms, and the stimulus (controlled upper airway obstruction) is similar to what occurs naturally in OSA.


In conclusion, a procedure for determining phenotypic traits in patients with OSA has been presented. The technique has been simplified considerably and has proven more feasible than the original approach. Although the procedure still has limitations, it nevertheless provides a potential framework for phenotyping patients in a clinical setting. Such a technique could possibly be used in the future to identify individuals for selected therapy.


Support for this study was provided by National Heart, Lung, and Blood Institute Grants 5R01 HL-048531-16, R01 HL-085188, R01 HL-090897, R01 HL-102321, K24-HL-093218, and P01 HL-095491; and by American Heart Association Grants 0840159N and 0575028N.


Andrew Wellman is a consultant for Philips Respironics, Sova, and Apnex. Robert Owens is a consultant for Apnex. Atul Malhotra receives research and/or consulting income from Philips Respironics, Apnex, Pfizer, SGS, SHC, and Apnicure, but has relinquished all outside personal income since May 2012. David White is Chief Scientific Officer for Apnicure.


The authors acknowledge Lauren Hess, Erik Smales, Pamela DeYoung, and Alison Foster for technical assistance with the studies.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
  29. 29.
  30. 30.
  31. 31.
View Abstract