Developing Nurse-Accessible Hypertension Prediction Tools for Low-Income Populations: A Comparative Analysis of Machine Learning Algorithms With SHAP Interpretation

Abstract

Aim

The aim of this study is to develop and compare machine learning algorithms for hypertension prediction in low-income populations, with emphasis on model interpretability for nursing implementation in resource-limited settings.

Methods

This retrospective cross-sectional study analysed data from seven iterations of NHANES (2005–2018) focusing on low-income populations. After LASSO regression identified eight key predictors, eight machine learning models were developed and evaluated using ROC curves, calibration plots and decision curve analysis, with SHAP methodology applied for interpretation.

Results

Among 12 506 participants, 39.96% had hypertension. Logistic regression and neural networks both achieved the highest discriminative ability (AUC = 0.853). SHAP analysis identified age as the most influential predictor, followed by waist circumference and diabetes status. A clinical nomogram with three-tier risk stratification (< 30%, 30%–60% and > 60%) was developed for nursing assessment.

Conclusion

Neural network models with SHAP interpretation achieved optimal hypertension prediction (AUC = 0.853) while maintaining clinical transparency essential for nursing practice. The resulting nurse-accessible nomogram with a visual scoring system supports evidence-based screening in low-income populations, pending external validation in clinical settings.