The Where and the Why of Obstructive Sleep Apnea

I don’t think much about anatomy during routine clinical visits. Terms like “velopharynx” and “lateral pharyngeal walls,” deeply etched in memory by years of training and reading, seldom find their way into my notes. Admitting this is a little embarrassing: I’m a sleep doctor, one whose practice is overwhelmingly dominated by obstructive sleep apnea (OSA). The upper airway, in all its contours, is my bread and butter, isn’t it?

In my defense, “don’t do the test unless it will change management” (1) left an impression as indelible as any anatomic term. The adage occurs to me with almost every head and neck examination, especially during the ridiculous, mirrored pantomime of a telehealth visit. (“Turn your head this way… no, that way.”) Why bother with this? Ostensibly, a good examination offers clues on the specific site of upper airway collapse—the “where” of OSA. But after justifying treatment with the “what” of OSA (the apnea–hypopnea index, the symptoms, the potential long-term sequelae), the “where” hardly matters. Continuous positive airway pressure (CPAP) distributes, like any pressure, throughout a given container, indifferent to structural details. If CPAP doesn’t discriminate, why should I?

Well, because CPAP is not the only OSA therapy, of course, and the “where” might point toward or away from non-CPAP alternatives. If, for example, I could exclude complete concentric collapse at the level of the soft palate (CCCp), I might have more confidence recommending oral appliance therapy (2) or hypoglossal nerve stimulation (3). Alas, the presence or absence of CCCp is difficult to predict by history and physical examination alone. Body mass index (4) and oropharyngeal findings (5) might vaguely suggest CCCp, but confidence requires specialized, semi-invasive procedures, principally drug-induced sleep endoscopy (DISE) (6). My physical examinations are therefore of little consequence to my management decisions, and I’m left with a nagging sense they’re performative, a kind of kabuki: Yes, I am a diligent, attentive physician. The mutterings of “scalloped tongue” and “mild overjet” testify to my thoroughness. How anticlimactic when the visit ends with the usual recommendation for CPAP and a tepid “You could try an oral appliance. It might work.”

Complementing the history and physical examination with a better sense for the “where” of OSA is a mission for precision sleep medicine: the process of developing and implementing metrics for choosing the right treatment for a given patient (7). But precision sleep medicine has lately focused elsewhere, on the “why” of OSA, the functional mechanisms or “endotypes” underlying apneas and hypopneas: increased collapsibility, low arousal threshold, high loop gain, and impaired muscular compensation. Methods like phenotyping using polysomnography (PUP) (8) are partly responsible, the focus the product of excitement that functional endotypes might shed their need for specialized CPAP dial-down studies (9). But what of structural endotypes? Might they also shed their need for DISE?

In this issue of AnnalsATS, Op de Beeck and colleagues (pp. 1567–1578) extend their work on the “where” of OSA through an investigation of overlap with the “why” (10). Previously (11), the authors compared airflow characteristics among 182 patients with OSA who underwent DISE and in-laboratory attended polysomnography, demonstrating that breath shapes may predict collapse site. For their new study, the authors performed a secondary analysis of the same cohort, this time comparing PUP-estimated functional endotypes among individuals with four structural sites of complete upper airway collapse: the soft palate (i.e., CCCp), tongue base, lateral pharyngeal walls, and epiglottis. Using a combination of categorical univariate tests and multivariate regression analyses, those with a given collapse site were compared with those without the given collapse site. In the univariate comparisons, CCCp was associated with greater PUP-estimated collapsibility, lateral wall collapse with reduced muscle compensation, and epiglottic collapse with greater muscle compensation. In the more complex, multivariate regressions, the authors assumed causal models in which the effect of a given PUP-estimated endotype on collapse site is mediated by the other three PUP-estimated endotypes. For example, arousal threshold, loop gain, and muscle compensation were assumed to mediate the effect of collapsibility (i.e., flow at eupneic ventilatory drive [V_passive]) on CCCp. That the effect of V_passive on CCCp remained significant after these adjustments was taken as evidence that the correlation of V_passive with CCCp explains something beyond V_passive’s known correlations with the other endotypes (12). That is, collapsibility has some “direct” relationship with CCCp beyond what might be explained by “indirect” relationships with loop gain, arousal threshold, and muscle compensation.

The study is an important early step in bringing the “where” and the “why” of OSA together. The univariate findings provide support for intuitions with limited direct evidence, for example that CCCp is associated with increased upper airway collapsibility. At first glance, this finding is muddied by the lack of an association between ventilation at nadir drive (V_min) and CCCp. V_passive and V_min are PUP-estimated collapsibility metrics, and, of the two, V_min correlates more closely with the reference standard pharyngeal critical closing pressure (13). However, the study included only patients with complete collapse at some airway level; individuals with partial collapse but no site of complete collapse were excluded. V_min is ventilation at the lowest levels of ventilatory drive. If low drive often implies low upper airway muscle dilator activity (14), finding little difference in V_min among individuals whose airways completely collapse is unsurprising. Zero flow at the tongue base is the same as zero flow at the soft palate. However, this defense highlights a limitation to generalizability: it is not clear how these findings apply to patients with only partial upper airway collapse, a potentially substantial proportion of patients with OSA.

The implications of the multivariate regression findings are more perplexing. The “mutual adjustment” approach is understandable in light of the difficulty of disentangling interactions among endotypes (15), and the authors deserve commendation for explicitly and transparently stating their assumptions (15). However, judging the correctness or incorrectness of these assumptions would require longitudinal data at multiple time points. Whether cause precedes mediator precedes effect is impossible to determine in a cross-sectional dataset, and the regressions ultimately amount to correlations across multiple dimensions, with the causal directions unclear. The authors (rightly) interpret their findings in the context of prior physiological knowledge rather than altering that knowledge in light of their findings.

Clinical meaning is especially elusive. Consider, for example, the nonsignificant relationship of arousal threshold with CCCp in the univariate analysis vis-à-vis the significant coefficient for the same relationship with adjustment for the other functional endotypes in the multivariate analysis. Such “reversal” might represent more than a statistical oddity (17), but readers should avoid the mistake of concluding that a low arousal threshold means CCCp is more likely, a potential takeaway of an unconsidered glance at figure 3. A more accurate synthesis would be “If my patient had the exact same collapsibility, loop gain, and muscle compensation, but the arousal threshold were lower, CCCp would become more likely.” I lack the imagination to parlay such a mouthful into something clinically actionable.

There is therefore no easy “PUP says this, so anatomy is that” conclusion. But any expectation that precision sleep metrics will one day dictate treatment decisions is likely mistaken. My teachers didn’t harp on anatomy because of some machine-like connection between findings and treatment; they did it to make me a better observer. I haven’t given up on the physical examination because, if I didn’t look, I might forget how to see. So the study by Op de Beeck and colleagues supports a look at the “why” of OSA: it might influence our thinking on the “where.” But we will still have to think.