Author(s): JOYCE, K., ASHDOWN, K., TRENDER, W., CLARKE, S., SIMPKINS, C., DELAMERE, J., BRADWELL, A.R., LUCAS, S.J.E., Institution: UNIVERSITY OF BIRMINGHAM, Country: UNITED KINGDOM, Abstract-ID: 1619

INTRODUCTION:
Acute mountain sickness (AMS) affects a large proportion of individuals ascending to high altitude each year. If left untreated, AMS can progress to serious illness (e.g., high-altitude cerebral oedema), which can be fatal or require emergency rescue. Despite this well-established observation and outcome, predicting AMS occurrence and severity for an individual prior to the onset of symptoms remains elusive. The objective of this study was to collect nightly pulse oximetry data during an ascent to 4554m (Margherita Hut) and, together with a classification model trained on data from a previous ascent to 4800m, utilise it to predict AMS severity at 4554m.
METHODS:
Twenty lowlanders (5 females) completed a 4-day ascent to 4554m where they spent two nights. Oximetry data were recorded continuously each night using wrist-worn pulse oximeters (WristOx2, Nonin Medical). Clinical evaluation was performed daily and Lake Louise Scores (LLS) collected each morning (AMS-positive: LLS greater than or equal to 3 with headache). AMS predictions were made using a classification model (Weighted K-Nearest Neighbour) that was developed (Kruskal-Wallis-based feature selection) and trained (MATLAB, 2022a, Mathworks) on data from a previous field expedition to 4800m. Sensitivity, specificity, and accuracy of the model were assessed by comparing model predictions against the LLS and clinical classifications (AMS-positive or AMS-negative). Similarly, model predictions based on overnight data from 3647m were compared to the occurrence of AMS at 4554m.
RESULTS:
Ninety three out of 100 oximetry recordings were successful with 743.1hrs (avg. recording length: 8.0±1.0hrs) of data collected, and minimal artifact observed (1.1±1.3%). Across the two nights at 4554m the model predictions exhibited: sensitivity: 36%; specificity: 92%; accuracy: 54%, for LLS classification, and sensitivity: 57%; specificity: 91%; accuracy: 76% for clinical classification. From the overnight data collected at 3647m, model predictions exhibited sensitivity: 57%; specificity: 83%; accuracy: 65% for LLS classification at 4554m, and sensitivity: 60%; specificity: 78%; accuracy: 65% for clinical classification at 4554m.
CONCLUSION:
Occurrence of AMS at 4554m can be predicted from overnight oximetry at 3647m using the present model. However, further research is required to optimise the model’s sensitivity, specificity, and accuracy for greatest impact and best real-world application. Future research should revaluate model predictions among a larger ascending cohort and also explore alternative overnight oximetry features for inclusion, as well as features from other sources (e.g., actigraphy, facial imaging).