Development and validation of a machine learning based early warning scoring system for high altitude polycythemia

Area of Science:

• Altitude Medicine
• Cardiovascular Health
• Machine Learning in Healthcare

Background:

• High-altitude polycythemia (HAPC) poses health risks to lifelong high-altitude residents.
• A need exists for lifestyle-focused risk stratification tools for HAPC in this population.
• Current risk assessment methods may not adequately address modifiable lifestyle factors relevant to high-altitude dwellers.

Purpose of the Study:

• To develop and validate a novel machine-learning predictive scoring system for HAPC.
• To utilize readily modifiable lifestyle variables for risk prediction in high-altitude populations.
• To create a tool for early identification and intervention of HAPC.

Main Methods:

• A cohort of 1,089 individuals residing at high altitude (≥4,500 m) was studied.
• 82 candidate variables were analyzed and reduced to seven key lifestyle predictors using LASSO.
• Logistic regression, XGBoost, and random forest models were trained and compared via 10-fold cross-validation.

Main Results:

• Logistic regression demonstrated the best performance with an AUC of 0.848, sensitivity of 0.81, and specificity of 0.79.
• Independent predictors of HAPC included low SpO2 (< 83%), male sex, age ≥50 years, smoking, hypertension, higher BMI, and lower tea consumption.
• These factors collectively contribute to a predictive model for HAPC risk.

Conclusions:

• The developed scoring system empowers frontline health workers in remote, high-altitude areas.
• It enables rapid identification of high-risk individuals for HAPC.
• Initiating low-cost lifestyle interventions can curb chronic altitude-related illnesses and improve quality of life.