Table of Contents
In South Africa (SA), stroke is the second highest cause of mortality and disability. Little is known about the importance of stroke predictors in SA. Identification of stroke predictors using appropriate statistical methods can help formulate appropriate health programs and policies aimed at reducing the stroke burden. Variable selection of predictors has concentrated mainly on stepwise technique which ignores stochastic errors, yet least absolute shrinkage and selection operator (LASSO) is more appropriate than using stepwise. The LASSO detect significant predictors from a pool of study variables. This study aims to identify and quantify important predictors through LASSO logistic regression (LLR) analysis. Methods: LLR and multivariate logistic regression were employed to establish a prediction model, we selected the most informative stroke predictors as well as examine the association of each predictor with stroke. Hospital-based data of 35,730 stroke cases were retrieved from selected private and public hospitals between January 2014 to December 2018 was used. Results: Of the 35,730 stroke cases, 22,183 were diabetic. The important predictors were hypertension, diabetes, the female gender, higher age groups, black and coloured races. The risk of stroke was found to increase in people with hypertension, and diabetes. The odds for females developing stroke was approximately 15% higher than males. The OR of black people developing stroke when compared to whites was 6 fold higher. Conclusion: The identified predictors can be used to raise awareness of modifiable predictors. Modelling predictors using the LASSO model could be beneficial for addressing the stroke burden in SA.
Keywords: Stroke, predictors, ordinary logistic regression, LASSO logistic regression, South Africa