Table of Contents
Stroke is the second leading cause of death and long-term disability in South Africa (SA). Yet little is known about modelling the modifiable and non-modifiable predictors in SA. The study aims to identify and quantify the effect of modifiable and non-modifiable stroke predictors using classical and Bayesian quantile statistics. Analyses of stroke predictors have mainly used logistic regression, yet using quantile regression (QR) is more appropriate than using mean regression in that QR provides flexibility to analyse stroke predictors corresponding to quantiles of interest. The mean regression models often miss critical aspects of the relationship that may exist between stroke and its predictors. QR analysis was used to model the effects of each predictor on stroke. Bayesian QR models were fitted for predictors. Both quantile and mean regression models were fitted and their estimates compared. Methods: A cross-sectional hospital-based study was used to identify and quantify stroke predictors in South Africa using 35,730 individual patient data retrieved from selected hospitals between January 2014 and December 2018. Results: The dominant stroke predictors were diabetes, hypertension, heart problems, the female gender, higher age groups and black-race. Higher age groups, the female gender and black-race had bigger effect on stroke distribution in the lower than upper quantiles. Conclusions: We confirmed that QR fits better than mean regression when modelling stroke predictors. The increasing stroke burden in SA needs urgent attention. The stroke predictors identified in the study can be used to raise awareness on modifiable predictors, promotion campaigns for healthy dietary choices.
Keywords: Stroke, modifiable, non-modifiable, predictors, Bayesian quantile regression model, South Africa