Introduction: The age-related dynamics of blood pressure (BP) arise from complex, often concurrent interactions among multiple factors (e.g., sex, type-2 diabetes mellitus [T2DM], body mass index [BMI]), making it challenging to isolate individual variable effects. Disentangling factor's contribution to BP trajectories across the lifespan remains a challenge.
Aim: Machine learning (ML) algorithms were applied to data from 219 individuals from a publicly available dataset to model age-related trends in systolic and diastolic BP, using age, sex, BMI, heart rate, and T2DM as predictors.
Methods: Five regression models (linear regression, random forests, support vector machines, gaussian process regression [GPR], and neural networks) were tested. The best-fitting models capturing complex predictor-target relationships were used to simulate systolic and diastolic BP trajectories under customized scenarios, independently varying sex, BMI, and T2DM to assess isolated effects.
Results: The squared exponential GPR yielded the best predictions for systolic BP, while the Matern 5/2 kernel performed best for diastolic BP. Systolic BP increased with age, with steeper trends at higher BMI. Women had lower systolic BP in early and mid-adulthood, but values surpassed men's in older age, especially with T2DM. Diastolic BP rose until midlife, then declined in both sexes. Women showed a similar crossover pattern, attenuated by T2DM, particularly at higher BMI.
Conclusion: ML simulations from a static dataset assessed individual factors' contributions to BP trajectories, producing results consistent with empirical evidence (e.g., greater T2DM impact on BP dynamics and faster age-related BP rise in women than men) and highlighting the potential for counterfactual analyses.
扫码关注我们
求助内容:
应助结果提醒方式:
