January, 2nd, 2023
Hypertension is a major global health burden that affects all age groups, causing 8.5 million deaths annually and contributing to 7% of disease burden worldwide. Identifying individuals at high risk of developing hypertension is crucial for implementing early preventative strategies and treatment. Risk prediction models that estimate the likelihood of developing hypertension using demographic and clinical data are widely used to identify individuals at high risk.
Traditionally, regression-based methods such as logistic regression and Cox regression have been used to develop prediction models. However, in recent years, machine learning algorithms have gained popularity as an alternative approach due to their ability to model nonlinear relationships and improved overall prediction accuracy.
In a recent study, machine learning algorithms were compared with a conventional Cox proportional hazards model for predicting hypertension incidence in a Canadian population. The study analyzed data from 18,322 participants and 24 candidate features from the Alberta's Tomorrow Project to develop prediction models. Five machine learning algorithms were developed: penalized regression (Ridge, Lasso, and Elastic Net), random survival forest, and gradient boosting. The predictive performance of the models was assessed using the C-index, with the performance of the machine learning algorithms found to be similar to that of the conventional Cox model.
Important features associated with each model were also identified. In general, age, sex, body mass index, and systolic blood pressure were identified as the most important features in predicting hypertension incidence.
Overall, the findings of this study suggest that machine learning algorithms can be used effectively for predicting hypertension incidence, with performance similar to that of conventional regression-based models. Further research is needed to fully understand the potential of machine learning algorithms in predicting hypertension and other health outcomes.