Development and validation of an AI approach for hyperbilirubinemia assessment from routine haematology in high-altitude settings
Related Experiment Video
Updated: Oct 31, 2026
A Point-of-Care Method with Integrated Decision Support Tool to Estimate Anemia at Population Level
Published on: January 19, 2024
Yulong Liu1,2, Mao Huang3, Juanjuan Mou3
1Department of Laboratory Medicine, The Affiliated Dazu's Hospital of Chongqing Medical University, Chongqing, China.
Background:
Hypobaric hypoxia at extreme altitudes induces compensatory erythropoiesis, alters bilirubin metabolism, and increases the risk of high-altitude hyperbilirubinemia (HBIL). Conventional serum bilirubin screening in these regions is frequently constrained by limited medical infrastructure. This study aims to develop a machine learning pipeline that uses universally accessible routine haematological parameters to infer serum bilirubin status and stratify disease severity.
Methods:
We conducted a retrospective cohort study of permanent high-altitude residents (>2,900 m) to develop and validate diagnostic models using routine haematological parameters. We employed a Tabular Prior-Data Fitted Network for primary diagnosis and clinical severity stratification (latent vs. overt jaundice). We evaluated model performance using the area under the receiver operating characteristic curve (AUROC) and validated the approach in an independent temporal hold-out validation cohort. We interpreted feature importance using SHapley Additive exPlanations (SHAP).
Results:
The model exhibited strong discriminative capacity, achieving an AUROC of 0.918 for primary diagnosis and 0.888 for disease severity stratification in the internal validation set. It maintained predictive efficacy in the temporal hold-out validation cohort, with AUROC values of 0.923 and 0.905, respectively. Interpretability analysis revealed that the magnitude of compensatory erythrocytosis (Hb and Hct) was primarily associated with the initial onset of HBIL. Furthermore, erythrocyte morphological heterogeneity (RDW-SD) distinguished the transition to overt jaundice.
Conclusions:
This study establishes a scalable, artificial intelligence-driven framework for screening and risk-stratifying high-altitude HBIL using routine blood parameters. This approach provides a practical, low-cost strategy for proactive clinical management in resource-constrained extreme environments.