Effects of the Combination of Atomoxetine and Oxybutynin on OSA Endotypic Traits

Abstract

Background

We recently showed that administration of the combination of the noradrenergic drug atomoxetine plus the antimuscarinic oxybutynin (ato-oxy) prior to sleep greatly reduced OSA severity, likely by increasing upper airway dilator muscle activity during sleep. In patients with OSA who performed the ato-oxy trial with an esophageal pressure catheter to estimate ventilatory drive, the effect of the drug combination (n = 17) and of the single drugs (n = 6) was measured on the endotypic traits over a 1-night administration and compared vs placebo. This study also tested if specific traits were predictors of complete response to treatment (reduction in apnea-hypopnea index [AHI] > 50% and < 10 events/h).

Methods

The study was a double-blind, randomized, placebo-controlled trial. The arousal threshold, collapsibility (ventilation at eupneic drive [Vpassive]), ventilation at arousal threshold, and loop gain (stability of ventilatory control, LG1), were calculated during spontaneous breathing during sleep. Muscle compensation (upper airway response) was calculated as a function of ventilation at arousal threshold adjusted for Vpassive. Ventilation was expressed as a percentage of the eupneic level of ventilation (%_eupnea). Data are presented as mean [95% CI].

Results

Compared with placebo, ato-oxy increased Vpassive by 73 [54 to 91]%_eupnea (P < .001) and muscle compensation by 29 [8 to 51]%_eupnea (P = .012), reduced the arousal threshold by –9 [–14 to –3]% (P = .022) and LG1 by –11 [–22 to 2]% (P = .022). Atomoxetine alone significantly reduced arousal threshold and LG1. Both agents alone improved collapsibility (Vpassive) but not muscle compensation. Patients with lower AHI, higher Vpassive, and higher fraction of hypopneas over total events had a complete response with ato-oxy.

Findings

Ato-oxy markedly improved the measures of upper airway collapsibility, increased breathing stability, and slightly reduced the arousal threshold. Patients with relatively lower AHI and less severe upper airway collapsibility had the best chance for OSA resolution with ato-oxy.

Our recent research showed that administration of the combination of atomoxetine and oxybutynin (ato-oxy) reduced the apnea-hypopnea index (AHI) by > 60% in a group of 20 patients with OSA.¹ The mechanism hypothesized for this effect is that noradrenergic (atomoxetine) and antimuscarinic (oxybutynin) agents can increase the central stimulation of motoneurons controlling the upper airway dilator muscles, such as the genioglossus.^2,3 This effect has been confirmed by the fact that an approximately threefold increase in genioglossus responsiveness to esophageal pressure (P_es) swings was measured with the drugs compared with placebo in these same subjects. Despite the group data showing an important reduction in AHI when the drugs were administered in combination, no effect on OSA severity was detected when the drugs were administered alone. The mechanisms underlying this observation remain unclear.

For an OSA intervention to be effective, it is expected to have a favorable effect on one or more of the following pathogenic traits⁴: (1) a collapsible pharyngeal airway, represented by low ventilation at normal ventilatory drive (Vpassive); (2) a reduced pharyngeal muscle responsiveness or compensation, represented by a failure to increase ventilation as ventilatory drive increases during sleep; (3) an oversensitive respiratory control system (loop gain [LG]), namely an abnormal ventilatory drive response to a prior reduction in ventilation; and (4) a low ventilatory drive that triggers arousal from sleep (low arousal threshold).⁵ The possibility of measuring an individual patient’s OSA endotypic traits during spontaneous breathing at atmospheric pressure6, 7, 8 enables us to investigate the effects of the drugs (administered alone or in combination) on these important OSA mechanisms. Importantly, the separate effects of these agents on OSA pathophysiologic traits have never, to the best of our knowledge, been assessed in patients with OSA. Accordingly, the first aim of the current study was to quantify the effect of the drug combination and the drugs administered alone for one night on the endotypic traits in a subgroup of patients who performed the original study. An esophageal catheter was used for the estimation of the ventilatory effort or drive. As a secondary aim, we tested if specific endotypic traits can identify the subgroup of patients with OSA who would have the best chances for complete OSA resolution using this combination of drugs. Indeed, approximately one-third of the subjects in the original study still had an AHI > 10 events/h and thus were not completely treated by the drug combination. Based on previous research testing the noradrenergic drug desipramine, we hypothesized that the patients with a milder anatomic defect and minimal upper airway muscle compensation⁹ would have the best improvement with ato-oxy treatment.

Patients and Methods

The effect of the drug combination on endotypic traits was a prespecified outcome of the protocol¹⁰ whose primary outcome (AHI) was previously published.¹ Analysis was not performed in patients who declined measurements of P_es. The participants underwent a double-blind, randomized, crossover trial performed at the Center for Clinical Investigation of Brigham and Women’s Hospital in which two overnight sleep studies were performed approximately 1 week apart: a placebo night and an atomoxetine 80-mg plus oxybutynin 5-mg (ato-oxy) night (Fig 1). After the unblinding of the study, patients who reported an AHI > 10 events/h on placebo night (n = 15) were asked to come back for two additional nights in which atomoxetine and oxybutynin were administered alone in random order. Thus, a subgroup of participants underwent four study nights: placebo, ato-oxy, atomoxetine alone, and oxybutynin alone.

Figure 1.

Consolidated Standards of Reporting Trials diagrams of the clinical trial testing ato-oxy (A) and of the trial testing the effects of the drugs administered alone (B). Results of the primary outcome analysis were previously published in Taranto-Montemurro et al.¹ ato-oxy = combination of atomoxetine and oxybutynin.

Detailed descriptions of the study design, methods, and equipment used have been reported elsewhere.¹ Briefly, in addition to the standard clinical montage for polysomnography,¹¹ flow was assessed with a pneumotach (Hans-Rudolph; Validyne) attached to a sealed oronasal mask (AirFit; ResMed), with the exhaust port (Whisper Swivel; Philips Respironics) attached to the pneumotach. Ventilatory drive was estimated from the P_es swings, which were measured by using a small, flexible, pressure-tipped catheter positioned in the lower third of the esophagus (Millar Instruments). Electromyographic activity from the genioglossus muscle (EMG_GG) was measured by using two intramuscular electrodes to create a bipolar recording.¹² Ato-oxy or placebo capsules were administered in random order 30 min prior to bedtime.

Data Processing

Respiratory Events and Genioglossus Activity

Apneas, hypopneas, and arousals were scored by using standard American Academy of Sleep Medicine guidelines¹³ by a registered polysomnography technician blinded to the treatment allocation. Hypopneas were defined as a reduction in flow ≥ 30% from baseline, lasting at least 10 s, and associated with an arousal from sleep or an oxyhemoglobin desaturation ≥ 3%. Genioglossus muscle responsiveness to progressively greater P_es swings during spontaneous breathing was calculated as previously described.¹

Arousal Threshold Measurement

P_es swings were quantified based on the nadir pressure minus the level at the start of inspiration. Similar to previous physiological studies,⁷ P_es swings immediately prior to arousal onset were identified and included in the analysis if they were preceded by at least three progressively greater P_es swings in the presence of flow-limited breaths or apnea. Arousals that did not meet these criteria were otherwise considered spontaneous (nonrespiratory) and were excluded from the analysis. The arousal threshold for each patient was defined as the median of the P_es swings before arousals during non-rapid eye movement (NREM) sleep.

Pharyngeal Collapsibility Measurements

Similar to studies from our laboratory,^6,7 a breath-by-breath “endotype plot” or “endogram” of ventilation vs ventilatory drive during sleep was constructed using P_es swings as a surrogate of ventilatory drive. Ventilation was expressed as a percentage of the eupneic level of ventilation (%_eupnea). Eupneic ventilation was calculated as the average ventilation for the whole night, including sleep epochs and arousals but not wakefulness epochs.⁷ Eupneic pressure swings were calculated as follows: (1) breaths recorded during wakefulness (≥ 4 breaths away from sleep) were identified; (2) the ratio of ventilation to P_es swing was measured for each awake breath; (3) the median ratio Y = ventilation/P_es was taken to reflect normal respiratory mechanics (ie, the amount of ventilation produced for a given P_es swing when the airway was open); and (4) eupneic pressure swings (eupneic ventilatory drive) were then calculated as [eupneic ventilation]/Y.⁶

Pharyngeal collapsibility was defined as the ventilation during sleep at eupneic ventilatory drive when the pharyngeal muscles are relatively passive (Vpassive); a more collapsible airway is captured by a lower ventilation. Ventilation at arousal threshold (Vactive) was defined as the level of ventilation at maximum drive (ie, ventilatory drive at the arousal threshold).^6,7,14 Analyses were completed before the unblinding of allocation order.

LG Measurement

The dynamic LG can be quantified as the ventilatory “response” to a ventilatory “disturbance” such as the reduced ventilation during apneas and hypopneas. LG was calculated from polysomnographic flow signals as the response to a one cycle/min disturbance (LG1), as previously described.⁸

Statistical Analysis

Mechanisms of Therapy

Two models were developed to assess the effects of treatment on each trait (and AHI). First, the effect of the ato-oxy combination vs placebo was modeled by using a linear mixed effects model analysis, with treatments as fixed effects and subjects as a random effect. Second, we deconstructed the effects of the combination by modeling separate additive and interaction effects (when evident) of the separate drugs; interaction terms were included only when physiologically significant trends were evident (to maximize statistical power for separate drug effects).

Effects on Vpassive (collapsibility) was modeled by using a sigmoidal transformation function (slope of 1 at Vpassive = 50%) to handle the known⁶ floor and ceiling effects; changes in collapsibility using our method are only linearly related to underlying collapsibility in the flow-limited range between Vpassive = 0% (apnea) and Vpassive = 100% (open airway). Effects on muscle compensation were estimated by modeling Vactive (same sigmoidal function) while adjusting for Vpassive. Effects on Vpassive and muscle compensation were not back-transformed, to estimate the underlying effects on physiology. Effects on LG and arousal threshold were modeled by using simple linear models. Effects on AHI were modeled by using square root transformation to normalize the skewed distribution; results were back-transformed for presentation.

Determinants of Response to Therapy

Bivariate and multivariate linear regression were conducted by using the residual AHI on ato-oxy as a dependent variable and Vpassive, fraction of hypopneas (hypopnea index/total AHI), BMI, neck circumference, baseline AHI, and muscle compensation (calculated as Vactive adjusted for Vpassive) as independent variables. These variables were chosen as indicators of unfavorable upper airway anatomy or lower muscle compensation at baseline according to our hypothesis. Age and sex were also included as possible confounding factors. Exploratory analysis assessing bivariate linear regression between residual AHI and %change in AHI (dependent variables) and all endotypic traits (independent variables) was also performed.

Statistical analyses were performed by using Matlab R2018a (MathWorks), GraphPad Prism 6.0 (GraphPad Software), and SPSS (IBM SPSS Statistics, IBM Corporation). A P value < .05 was considered statistically significant. Data are presented as mean [95% CI].

Results

Patients

Eighteen of the 20 patients who completed the parent trial had P_es measurements and could be included in this study. One patient did not have sleep apnea on both nights (AHI < 1 event/h) and was excluded from the endotypic traits analysis. The main characteristics of this group of 17 subjects are reported in Table 1. No serious adverse events occurred. Complete data for this trial were previously published.¹ Six of the nine patients who performed the additional single-drug study had P_es measurements and were included in the current study. Their characteristics are reported in Table 2.

Table 1.

Characteristics of Patients Considered for Endotypic Traits Analysis in the Trial Testing the Drugs Taken in Combination (N = 17)

Female sex	2 (12)
Age, y	53 [50-60]
BMI, kg/m2	35.0 [30.2-40.8]
Neck circumference, cm	43.6 [40.0-44.8]
Waist circumference, cm	113.0 [103.5-130.5]
OSA treatment
CPAP compliant	4 (24)
CPAP noncompliant	5 (29)
Oral appliance	1 (6)
None	7 (41)
Comorbidities
Hypertension	7 (41)
Diabetes	2 (12)
Hypercholesterolemia	3 (18)

Data are presented as No. (%) or median [interquartile range].

Table 2.

Characteristics of Patients Considered for Endotypic Traits Analysis in the Trial Testing the Drugs Taken Alone (n = 6)

Female sex	2 (33)
Age, y	55 [47-61]
BMI, kg/m2	35.5 [32.4-38.9]
Neck circumference, cm	44 [42-44]
Waist circumference, cm	113 [106-122]
OSA treatment
CPAP compliant	2 (33)
CPAP noncompliant	3 (50)
Oral appliance	0
None	1 (17)
Comorbidities
Hypertension	3 (50)
Diabetes	1 (17)
Hypercholesterolemia	1 (17)

Data are presented as No. (%) or median [interquartile range].

OSA Severity and EMG_GG

Ato-oxy reduced OSA severity in 16 of the 17 patients included in the analysis. Sleep and OSA severity metrics are reported in Table 3. Genioglossus muscle responsiveness results for the 16 patients in whom EMG_GG was measured were previously reported elsewhere¹ and are listed in Table 3.

Table 3.

OSA Severity and Sleep On and Off the Study Drugs

Variable	Placebo	Ato-oxy	Median Change	P Value
AHI, events/h	31.2 [11.9 to 55.7]	7.6 [2.4 to 20.3]	–28 [–40.0 to –10.5]	< .001
%change			–65 [–88 to –52]
AHI NREM,a events/h	32.5 [8.5 to 55.7]	7.6 [2.4 to 20.3]	–22.0 [–40.0 to –7.6]	< .001
%change			–65 [–88 to –40]
AHI REM,a,b events/h	37.8 [25.4 to 56.3]	2.2 [0 to 6.6]	–31.3 [–54.1 to –18.5]	.007
%change			–92 [–95 to –83]
AI, events/h	5.3 [1.3 to 15]	0.3 [0 to 2.1]	–1.3 [–13.3 to –0.8]	< .001
Fhypopneas, %AHI	67.1 [44.3 to 86.8]	93.3 [69.0 to 100]	19.7 [–2.1 to 42.8]	.074
ODI 3%, events/h	17.1 [6.3 to 37.4]	3.7 [0.6 to 20.2]	–6.2 [–16.7 to –2.6]	< .001
Nadir Sao2, %	82 [78.5 to 90.5]	94 [88.5 to 95.5]	9 [4 to 16]	< .001
Arousal index, events/h	46.4 [22.2 to 61.6]	43.4 [33.5 to 52.8]	–6.1 [–15.7 to 8.3]	.35
Total sleep time, min	232 [162 to 326]	303 [237 to 321]	45 [–18 to 89.5]	.11
Sleep efficiency, %TIB	58 [41 to 83]	75 [63.5 to 86]	9 [–2 to 23.5]	.043
EMGGG, %baseline	2.2 [1.1 to 4.7]	6.3 [3.0 to 18.3]	2.7 [1.0 to 9.2]	< .001

Data are presented as median [interquartile range]. Baseline was calculated as the median Peak EMG_GG during non-REM sleep with non-flow-limited breaths and relatively small esophageal pressure swings (–11.0 [–17.6 to –6.8] cm H₂O). AHI = apnea-hypopnea index; AI = apnea index; Ato-oxy = combination of atomoxetine and oxybutynin; EMG_GG = electromyographic activity from the genioglossus muscle; F_hypopneas = fraction of hypopneas over total events; NREM = non-rapid eye movement sleep; ODI 3% = oxygen desaturation index, 3% desaturation; ODI = oxygen desaturation index; REM = rapid eye movement; Sao₂ = arterial oxygen saturation; TIB = time in bed.

^aAHI is calculated in supine position.

^bREM AHI was calculated only in eight patients in whom at least 10 min of REM sleep were available in both nights.

Endotypic Traits During NREM Sleep

Overall Effects of the Combination

Individual data of endotypic traits on placebo and ato-oxy are shown in Figure 2A for NREM sleep and Figure 2B for rapid eye movement (REM) sleep. Mixed effects model analysis was performed only for NREM sleep, and the results are shown in Figure 3 and Table 4. Compared with placebo, ato-oxy increased Vpassive by approximately 73%_eupnea, muscle compensation (Vactive adjusting for Vpassive) by approximately 29%_eupnea, reduced arousal threshold by approximately 8%, and lowered LG (LG1) by approximately 10%.

Figure 2.

Individual data showing the effect of ato-oxy on the endotypic traits assessed during spontaneous breathing during NREM (A) and REM (B). Vpassive and Vactive were increased with ato-oxy. The arousal threshold and LG were reduced on ato-oxy. LG of the respiratory system was calculated here at the frequency of one cycle/min. Lines indicate mean values. AHI = apnea-hypopnea index; NREM = non-rapid eye movement sleep; %_eupnea = percentage of the eupneic level of ventilation; REM = rapid eye movement sleep; Vactive = ventilation at the arousal threshold; Vpassive = ventilation at eupneic drive. See Figure 1 legend for expansion of other abbreviation.

Figure 3.

Group data showing the effects of atomoxetine, oxybutynin, and ato-oxy (in combination) on the endotypic traits in all patients during NREM sleep. Only the patients who performed the study with the P_es measurement were included in the analysis: n = 17 for placebo and ato-oxy in combination; n = 6 for atomoxetine and oxybutynin administered alone. The Consolidated Standards of Reporting Trials diagrams in Figure 1 also provide more details. Lines represent means, and the boxes indicate 95% CIs from mixed effects model analysis (Model 2, Table 3). Muscle compensation data illustrate the model-estimated underlying effects of interventions on Vactive adjusting for Vpassive (equivalent to Vactive minus Vpassive but accounting for floor/ceiling effects). ∗P < .05, ^†P < .01, ^‡P < .001 vs placebo. P_es = esophageal pressure. See Figure 1 and 2 legends for expansion of other abbreviations.

Table 4.

Mixed Effects Model for Effect of Atomoxetine, Oxybutynin, and Ato-oxy vs Placebo on AHI, Vpassive, Compensation, Arousal Threshold, and Loop Gain During NREM Sleep

Variable	Vpassive (%eupnea)	Muscle Compensation (%eupnea)	Arousal Threshold (cm H2O)	Loop Gain (Unitless)	AHI (Events/h)
Model 1
Constant	63 [43 to 83] P < .001	50 [23 to 77] P < .001	35 [28 to 42] P < .001	0.55 [0.49 to 0.60] P < .001	28 [20 to 40] P < .001
Ato-oxy (change from baseline)	73 [54 to 91] P < .001	29 [8 to 51] P = .012	–3 [–5 to –1] P = .022	–0.06 [–0.12 to –0.01] P = .022	–24 [–26 to –21] P < .001
Model 2
Constant	62 [46 to 78] P < .001	43 [21 to 64] P < .001	35 [28 to 42] P < .001	0.55 [0.50 to 0.61] P < .001	30 [21 to 40] P < .001
Atomoxetine (change from baseline)	38 [21 to 56] P < .001	–1 [–22 to +21] P = .96	–6 [–11 to –2] P = .003	–0.10 [–0.15 to –0.05] P < .001	–1 [–10 to 9] P = .64
Oxybutynin (change from baseline)	18 [4 to 34] P = .019	–3 [–21 to +14] P = .70	0 [–4 to +4] P = .87	0.04 [–0.02 to 0.09] P = .20	4 [–5 to 15] P = .30
Interaction (change from baseline)	…	26 [–7 to 59] P = .15	4 [–2 to 10] P = .23	…	–24 [–29 to –15] P < .001

Results are reported as mean [95% CIs]. Ato-oxy indicates the effect of treatment (combination of atomoxetine and oxybutynin) vs placebo. Due to a skewed distribution, AHI data values were square root transformed for analysis and back-transformed for presentation. Values for Vpassive do not represent observed data (presented in Figures 2 and 3) but rather the underlying collapsibility derived from a sigmoidal transformation function, to handle the ceiling effects previously described for these types of data.⁶ Values for Muscle Compensation were calculated from Vactive adjusting for Vpassive such that the effect shown is the additional effect on ventilation above Vpassive (thus representing pharyngeal compensation). Constant represents the placebo AHI or trait values. See Table 3 legend for expansion of abbreviations.

Separate Effects

Separately, atomoxetine and oxybutynin increased Vpassive by approximately 38%_eupnea and approximately 18%_eupnea, respectively, with no evidence of an interaction. Muscle compensation (Vactive adjusting for Vpassive) was not significantly increased by either of the separate drugs; a physiologically relevant but nonsignificant trend for interaction was present (approximately 25 %_eupnea), which would signify that both agents may be needed to improve muscle compensation. Atomoxetine alone reduced the arousal threshold by 6 cm H₂O, an effect that appeared larger than in the presence of oxybutynin (3 cm H₂O). Atomoxetine alone also lowered LG compared with placebo. Group data are reported in Figure 3.

Figure 4 shows the endotype plot of three patients representative of the different endotypic traits measured at baseline and during ato-oxy therapy.

Figure 4.

Endograms of three representative patients on placebo and on ato-oxy. A, A patient with a severe OSA characterized by a bad anatomy (low Vpassive, blue dot), moderate muscle compensation (Vactive, red dot > Vpassive), and low arousal threshold (orange line) on placebo shows an impressive improvement in Vpassive and Vactive on the drugs; arousal threshold is slightly reduced. B, The patient has a good anatomy (Vpassive approximately 90%) on placebo but bad muscle compensation (Vactive is much lower than Vpassive), meaning that he cannot activate the upper airway dilator muscles during sleep despite an increase in ventilatory drive to > 200%_eupnea (negative effort dependence with high arousal threshold). On the drug, the negative effort dependence is greatly reduced, and the ventilation remains close to eupneic values even when the ventilatory drive is high. C, Patient with severe anatomy (low Vpassive), high arousal threshold, and very good muscle compensation (Vactive is much higher than Vpassive) on placebo. Vpassive is highly increased on ato-oxy, and ventilation remains close to eupnea even at higher ventilatory drive levels. Shaded areas indicate the interquartile range. See Figure 1 and 3 legends for expansion of abbreviations.

Endotypic Traits in REM Sleep

Individual data of endotypic traits on placebo and ato-oxy are shown in Figure 2B for REM sleep. Only seven patients had REM sleep on both the placebo and the ato-oxy night and only four patients had REM sleep on at least one of the nights when the single drugs were administered. In this subgroup of patients, the effect of the combination on the endotypic traits in REM sleep was similar to the effect measured during NREM sleep, although the effect on Vactive was nonsignificant during REM, probably due to small sample (type II error). Given the limited amount of information available, the mixed effects model analysis was not performed for REM sleep.

Predictors of Successful Treatment

In the bivariate analysis, residual AHI was directly related to baseline AHI (r = 0.82; P < .001) and inversely related to fraction of hypopneas (numbers of hypopneas /total obstructive events [F_hypopneas ]) (r = –0.80; P < .001) and Vpassive (r = –0.67; P = .009). In the multiple regression analysis, only baseline AHI and F_hypopneas were significant independent factors included in the model (R² = 0.81; P < .001). Figure 5 shows individual data: eight of 14 subjects had a complete response to ato-oxy (50% reduction in AHI from placebo and AHI on drug < 10 events/h). These data show that patients with an AHI on placebo < 40 events/h, F_hypopneas > 65%, and Vpassive > 55% were associated with the best likelihood of complete response on ato-oxy.

Figure 5.

These plots show a significant direct relation between residual AHI (on ato-oxy) and AHI on placebo (left panel) and an inverse relation between residual AHI and the F_hypopneas (center panel). The right panel shows an inverse relation between the residual AHI and the ventilation when ventilatory drive is at eupneic level on placebo (Vpassive, right panel). Shaded blue areas include patients with a complete response on ato-oxy (> 50% reduction in AHI from placebo and AHI on drug < 10 events/h). Vertical dotted lines represent cutoff values between responders and nonresponders for each independent variable AHI: apnea-hypopnea index. Only 14 patients with P_es measurements and a baseline AHI > 10 events/h were included in this analysis. F_hypopneas = fraction of hypopneas (numbers of hypopneas/total obstructive events). See Figure 1 and 2 legends for expansion of abbreviations.

We next performed a bivariate linear regression analysis including each trait measured on placebo night as an independent variable and the percent change in AHI as a dependent variable. We found a trend for a direct relation between change in AHI and Vpassive (the higher the Vpassive, the higher the AHI reduction) (r = 0.49; P = .073) and an inverse relation between change in AHI and LG1 (r = 45; P = .079), suggesting that patients with highly collapsible airway and high LG may have minor improvements in sleep apnea with the medications. Given the limited sample size, these findings clearly need to be confirmed in larger trials.

To understand if the change in the physiology traits could explain the reduction in AHI, we also performed a bivariate regression analysis of the percent change in AHI (dependent variable) vs the change in each trait (Vpassive, Vactive, muscle compensation, LG, and arousal threshold) as independent variable. There was a significant, direct correlation between the change in Vpassive and the change in AHI (higher reductions in AHI were associated with a higher increase in Vpassive) (r = 0.57; P = .03). There was also a trend for a relation between the change in muscle compensation and the change in AHI so that the patient in which the muscle compensation was increased the most on ato-oxy, also had the best improvement in AHI (r = 0.47; P = .089). These findings are consistent with the hypothesis that the drug’s mechanism involves the improvement of upper airway muscle function.

Discussion

In this study, we performed a detailed physiological analysis on a breath-by-breath basis to assess the effect atomoxetine and oxybutynin taken alone and in combination have on the four endotypic traits, as previously described by our group.⁵ We also tried to identify which baseline endotypic traits best predicted who would completely respond to ato-oxy. The main findings of this study were as follows: (1) ato-oxy improved airway collapsibility (greater Vpassive), muscle compensation (Vactive adjusted for Vpassive), stabilized breathing (reduction in LG1), and slightly reduced the arousal threshold measured with P_es swings; (2) atomoxetine alone reduced the arousal threshold and LG1, and atomoxetine and oxybutynin alone increased Vpassive but did not affect Vactive or muscle compensation; and (3) upper airway “passive” anatomy, the fraction of hypopneas over total events, and AHI at baseline predicted the resolution of OSA to ato-oxy administration. Moreover, there was a trend for a lower change in AHI in patients with high LG.

These results confirm our hypothesis that a therapy targeting the upper airway muscles could be effective in patients with a mild to moderate anatomic defect, identified in this population by a Vpassive higher than 55% of eupneic ventilation and a predominance of hypopneas over apneas on the polysomnogram. This outcome is consistent with previous literature¹⁵ showing that targeting nonanatomic traits could be an effective strategy in patients with mild/moderate airway collapsibility. Nevertheless, we found that even some patients with a severely collapsible airway experienced an important reduction in OSA severity, as shown in the endogram of Figure 4A. In this figure, a patient with frequent apneas on placebo (Vpassive = 4.2%_eupnea; F_hypopneas = 23.4%) experienced only shallow hypopneas (Vpassive = 91%_eupnea; F_hypopneas = 98.2%) on the drugs.

F_hyponeas is an approximate measure of collapsibility that correlates with the critical collapsing pressure measured during sleep.¹⁶ Previously, a higher F_hypopneas was identified as a predictor of low arousal threshold, together with lower AHI and milder oxygen desaturations.¹⁷ This relation was mainly explained by the authors with the fact that patients whose anatomy is better (indicated by less severe airflow obstruction and thus higher F_hypopneas) should be more likely to have a nonanatomic factor causing OSA, such as a low arousal threshold. If these data can be confirmed in larger trials, F_hypopneas may become a key feature for the selection of patients for this treatment, as it is already available from the polysomnographic reports in all sleep clinics, together with the AHI. On the contrary, Vpassive, LG, and the other traits are derived from a detailed breath-by-breath analysis⁶ using algorithms not yet available in commercial software, and their use is currently limited to a few research laboratories. However, given the need for more personalized treatment options for patients with OSA, an accurate endotypic analysis could soon become a widespread clinical tool.

Contrary to our previous results collected with the adrenergic drug desipramine,⁹ baseline muscle compensation was not a predictor of better response with ato-oxy. Possible explanations for this difference could be related to the diverse pharmacologic profiles of these drugs or the different methods through which the muscle compensation was calculated (ie, during spontaneous breathing at atmospheric pressure in this trial vs CPAP dial-downs in the desipramine study⁹).

Atomoxetine caused a reduction in the arousal threshold by 6.2 [4.5 to 11.8] cm H₂O (P = .03) when administered alone, with this effect likely being related to the alerting effects associated with the central increase in norepinephrine. In the same six patients, the reduction in arousal threshold on ato-oxy was a nonsignificant 2.6 [0.9 to 5.8] cm H₂O (P > .5). These data suggest that the administration of oxybutynin could attenuate the alerting effect of atomoxetine through mechanisms that need to be investigated. Accordingly, previous literature reported that antimuscarinic agents administered at low doses have mild sedative effects¹⁸ and induce sleepiness.¹⁹ Moreover, and consistent with this hypothesis, it was shown that oxybutynin can improve sleep quality by reducing symptoms of nocturia.²⁰ The consequence of this finding is important because it suggests that oxybutynin could be replaced with a sedative to increase the arousal threshold, potentially avoiding problematic antimuscarinic side effects; this hypothesis warrants further investigations.

Atomoxetine was likely the main drug responsible for the ventilatory improvement, as can be seen in Figure 3. Vpassive was significantly augmented by 38%_eupnea (P < .001) (Table 4) with atomoxetine compared with placebo. It is important to note that Vpassive does not represent here ventilation when the upper airway muscles are completely passive (such as happens during optimal CPAP treatment); rather, it is the ventilation at relatively low ventilatory drive (and relatively low pharyngeal muscle activity) compared with the ventilation at the arousal threshold (Vactive), in which ventilatory drive and upper airway muscle activity are maximal during sleep. Interestingly, the further increase in Vactive from atomoxetine to ato-oxy could be due to an increase in arousal threshold (more time allowed during sleep for muscle recruitment) or to a synergistic effect of the drug combination on the upper airway muscles, because atomoxetine or oxybutynin alone did not show a trend for a Vactive or muscle compensation improvement according to our data. Atomoxetine might also be responsible for the reduction in LG1 by approximately 10%. The reason is not clear at this time, but it may be related to an increased central respiratory motor output contributing to breathing stabilization.²¹

The current study has several limitations. First, the relatively small number of patients involved in the additional study on the single drugs limits the generalizability of our findings related to the individual effects of each drug. More data need to be collected, especially to identify the effects on Vpassive or Vactive and muscle compensation. Second, it is unknown what the effects are when these drugs would be administered for several nights, rather than one night only. Third, the endotype analysis that we performed is different from previous endotype analyses performed with the CPAP drop technique, and therefore the results reported here related to endotypic traits are difficult to compare with those obtained previously in patients with OSA. However, this endotyping technique, which uses the esophageal catheter to estimate ventilatory effort, allows the contemporary measurement of the AHI and the endotypic traits for the whole night, thereby reducing the number of study nights for each subject.

Conclusions

We showed that ato-oxy opened the airway and increased ventilation at all levels of ventilatory effort. Although atomoxetine alone was probably responsible for an unwanted reduction in arousal threshold, it seemed to better improve ventilation and ventilatory stability compared with oxybutynin. Furthermore, we found that patients with a less collapsible airway had a complete recovery from OSA while taking ato-oxy. More data on a wider group of patients need to be collected to better understand the effects of the single drugs on endotypic traits and the characteristics of complete responders. This study suggests that estimates of the endotypic traits from the polysomnogram6, 7, 8^,17,22 may provide useful information in the clinical and research settings to broaden the treatment possibilities for patients with OSA.