Reading the wrist as a mechanical observer of the collapsing pharynx: biomechanical inference of upper-airway soft tissue loading from wearable signals

The pharynx is a collapsible soft-tissue conduit whose patency is a mechanical equilibrium between intraluminal loading and tissue elastance, yet the transmural pressure that governs its collapse is accessible only through invasive laboratory manometry. Here, we reframe wearable sleep monitoring as inference over mechanically motivated surrogates of pharyngeal loading. From a Starling resistor and tube law description of pharyngeal collapse, we derive an explicit mapping from each wearable modality—photoplethysmography, electrocardiography, accelerometry, oximetry, and respiratory effort—to the mechanical term it is expected to reflect, casting the sensor array as a distributed set of mechanical observers of one collapse event. Three innovations follow: self-supervised pre-training restricted to mechanically admissible augmentations, retained only when a perturbed view remains a physically realizable airway state; a redundancy-aware fusion that combines correlated observers through a Sharpe-ratio-maximizing, closed-form weighting over their information covariance; and a proposed cumulative load index, formulated in the shape of a classical injury criterion, that summarizes a night of recording as an uncalibrated exposure to obstructive loading rather than an event count; the index is introduced as a hypothesis, not as a validated mechanical dose. Across two polysomnographic cohorts spanning ages 18–94, the framework surpasses 13 baselines in collapse detection and in three-way loading regime classification—sustained, absent, or delayed-onset loading—in both cohorts and retains its advantage when observers are lost, both under controlled channel dropout and under zero-shot transfer with only the two observers shared by the two corpora.