Predicting high blood pressure using machine learning models in low- and middle-income countries

Area of Science:

• Public Health
• Epidemiology
• Machine Learning in Healthcare

Background:

• Noncommunicable diseases (NCDs) are a growing global health concern, with high blood pressure being a key risk factor.
• Effective management of high blood pressure is crucial for controlling NCDs worldwide.

Purpose of the Study:

• To develop and evaluate explainable machine learning models for predicting high blood pressure using global survey data.
• To identify key predictors of high blood pressure across diverse populations and geographical regions.

Main Methods:

• Utilized harmonized data from 57 countries participating in the STEPwise approach to NCD risk factor surveillance (STEPS) surveys.
• Trained and evaluated five machine learning models (logistic regression, k-NN, random forest, XGBoost, neural network) using an 80/20 train-test split.
• Assessed model performance using accuracy, precision, recall, and F1 score, with feature importance analysis.

Main Results:

• Key predictors of high blood pressure included age, weight, heart rate, waist circumference, and height.
• Model accuracy varied significantly across countries, ranging from 58.96% to 81.41%.
• Performance differences underscore the necessity for region- and country-specific modeling strategies.

Conclusions:

• Explainable machine learning models can aid in population-level screening and risk assessment for high blood pressure.
• The findings emphasize the importance of adapting predictive models to local contexts for effective NCD risk factor surveillance.
• This approach offers a valuable tool for resource-limited settings to manage high blood pressure and mitigate NCDs.