Background

Acute mountain sickness (AMS) affects a large proportion of individuals ascending to high altitude every year. While AMS is not life threatening, if left untreated it can progress to serious illness (e.g., high altitude cerebral oedema). Current assessment tools rely on subjective symptom reporting and do not enable prospective prediction. Objective, field-deployable methods for early risk stratification remain limited. This study evaluated whether physiological biomarkers derived from overnight pulse oximetry can predict subsequent AMS during ascent to Everest Base Camp (EBC).

Methods

A field study was conducted in the Khumbu region of Nepal. Healthy lowlanders (n = 139–144 across datasets) underwent overnight pulse oximetry recordings at Namche Bazaar (3,440 m) and Dingboche-level (4,371-4,730 m). A total of 125 physiological features were extracted from continuous, overnight oxygen saturation (SpO 2 ) and heart rate recordings. Supervised machine-learning models were trained to classify physician-diagnosed AMS at Lobuche (~4,940 m) using: 1) physiological biomarkers alone and 2) biomarkers combined with clinical and contextual metadata. Model performances were assessed using repeated stratified k-fold cross-validation, with area under the receiver operating characteristic curve (AUC) as the primary metric.

Results

Biomarker-only models demonstrated consistent above-chance discrimination at both altitudes (mean AUC 0.74 at both Namche and Dingboche-level). At lower altitude, predictive performance was achieved using a small set of physiological features, and inclusion of metadata did not improve model performance. At higher altitude, combining biomarkers with contextual variables improved discrimination (AUC 0.84 ± 0.08; sensitivity 0.73 ± 0.23). Selected features were primarily related to nocturnal oxygenation burden, desaturation dynamics, and heart rate variability.

Conclusion

Overnight pulse oximetry captures physiological signals associated with subsequent AMS development prior to symptom onset. These findings support the feasibility of objective, non-invasive risk stratification during early ascent. Integration of wearable monitoring with computational modelling may provide a complementary tool to support preventative decision-making along high-altitude trekking routes.