Back

Using Discrete- and Continuous-Time Machine Learning Models (Nnet, Coxnet, Glmnet) to Explore Sex and Age Differences in Stroke Prediction Among Hypertensive Individuals

MacCarthy, G.; Pazoki, R.

2025-08-26 cardiovascular medicine
10.1101/2025.08.22.25334217 medRxiv
Show abstract

IntroductionStroke is one of the leading causes of death and long-term disability globally. Several studies have investigated the incidence and predictors of stroke in the healthy population; only a few have specifically focused on stroke risk prediction among individuals with hypertension. Given that hypertension is the most common modifiable risk factor for stroke, this represents an important research area to study. Improving our understanding of stroke risk among hypertensive individuals has the potential to significantly improve preventative efforts and decrease the worldwide stroke burden. Material and MethodsThe current study involved 116,216 participants of European ancestry from the UK Biobank. We created stroke genetic liability using data from two predictive models using clinical, lifestyle, and genetic factors (stroke genetic liability) in the training set, namely a Cox proportional hazard (CoxPH) model, a penalized Cox proportional hazard model (CoxNet), a penalized logistic regression model (GLMnet), and a neural network (Nnet), to estimate time-to-event risk for stroke among hypertensive patients, older and younger hypertensive patients, and hypertensive males and females separately. We then assessed their performances in the testing set with the area under the curve (AUC) and Brier score (BS). ResultsThe CoxPH, CoxNet, and GLMnet achieved an equal AUC of 67.0% in hypertensive patients. All models demonstrated superior performance in men compared to women, with the CoxPH model achieving the best AUC of 68% in men. All models demonstrated superior performance in older patients compared to younger patients, with the GLMnet achieving the best AUC of 61% in this group. ConclusionIncluding stroke genetic liability in prediction models slightly improves stroke prediction for hypertensive men and older patients. The Cox proportional hazards (CoxPH) model outperformed the machine learning models in predicting stroke risk among hypertensive men.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.