Acute mountain sickness prediction: a concerto of multidimensional phenotypic data and machine learning strategies in the framework of predictive, preventive, and personalized medicine.

内容摘要

Acute mountain sickness (AMS) is a self-limiting illness, involving a complex series of physiological responses to rapid ascent to high altitudes, where the body is exposed to lower oxygen levels (hypoxia) and changes in atmospheric pressure. AMS is the mildest and most common form of altitude sickness; however, without adequate preparation and adherence to ascent guidelines, it can progress to life-threatening conditions. Due to the multi-factorial predisposition of AMS among individuals, identifying AMS biomarkers before high altitude exposure from multiple dimensions (e.g., clinical, metabolic, and proteomic markers) and integrating them to build an AMS predictive model enables early diagnosis and personalized interventions, which allows targeted allocation of medical resources, such as prophylactic medications (e.g., acetazolamide) and supplemental oxygen, to those who need them most and prevention of unnecessary complications. Consequently, predicting AMS utilizing biomarkers from multidimensional phenotypic data before high-altitude exposure is essential for the paradigm change in high-altitude medical research from currently applied reactive services to the cost-effective predictive, preventive, and personalized medicine (PPPM/3PM) in primary (reversible damage to health and targeted protection against health-to-disease transition) and secondary (personalized protection against disease progression) care. To this end, this study recruited 83 Han Chinese male volunteers and obtained clinical, proteomic, and metabolomic profiles for analysis before they ascended to high altitudes. The Mann-Whitney U test was used to identify clinical features distinguishing AMS from non-AMS. The proteomic and metabolomic features were concatenated and clustered to find co-expression modules associated with AMS. A machine learning model, Mutual Information-radial kernel-based Support Vector Machine-Recursive Feature Elimination (MI-radialSVM-RFE) was employed for biomarkers selection and AMS prediction. A molecular docking technique was used to select molecular biomarkers that can bind with Traditional Chinese Medicine (TCM) ingredients. Among 83 participants, 66 were selected for detailed analysis after quality control steps. Six protein-metabolite co-expression modules were identified as significantly associated with AMS. The MI-radialSVM-RFE model selected 12 biomarkers (two clinical features: systolic blood pressure (SBP) and peak expiratory flow (PEF); six proteins: Acyl-CoA synthetase long-chain family member 4 (ACSL4), immunoglobulin kappa variable 1D-16 (IGKV1D-16), coagulation factor XIII B subunit (F13B), prosaposin (PSAP), poliovirus receptor (PVR), and multimerin-2 (MMRN2); and four metabolites: 2-Methyl-1,3-cyclohexadiene, calcitriol, 4-Acetamido-2-amino-6-nitrotoluene, and 20-Hydroxy-PGE2) for the AMS prediction model. The model exhibited excellent predictive performance in both training (n = 66) and validating cohorts (n = 24) with AUCs of 0.97 and 0.94, respectively. Additionally, molecular docking analysis suggested PSAP and ACSL4 proteins as potential molecular targets for AMS prevention. This study advances high-altitude medicine by developing a predictive model for AMS using clinical, proteomic, and metabolomic data. The identified biomarkers linked to energy metabolism, immune response, and vascular regulation offer insights into AMS mechanisms. High-altitude predictive approaches should focus on implementing biomarker-driven risk screening using clinical, proteomic, and metabolomic data to identify high-risk individuals before high-altitude exposure. Preventive measures should prioritize pre-acclimatization protocols, tailored nutritional strategies and interventions guided by biomarker profiles, and lifestyle adjustments, such as maintaining mitochondrial health through proper nutritional strategies. The online version contains supplementary material available at 10.1007/s13167-025-00404-9.