Modelling predictors of stroke disease in South Africa: Bayesian binary quantile regression approach

Abstract

Stroke is currently the second prevalent cause of death and disability worldwide. South Africa (SA) is experiencing an epidemiological transition due to socio-demographic and lifestyle changes leading to an increase of non-comm-unicable diseases, which in turn may result in an upswing of stroke cases. Modifiable predictors cause most strokes. The purpose of this paper is to address two important gaps in the stroke disease literature that is identifying and modelling predictors of stroke and estimating linear quantile models when predictors are measured with error. Methods: A hospital-based cross-sectional study design was used to model the predictors of stroke incidences in SA. We estimated posterior marginal by integrated nested Laplace approximations (INLA) for latent Gaussian models. The main objective of this study is to assess the effects of predictors of stroke for different quantiles for adults stroke patients and to estimate linear quantile regression models when predictors are measured with error. We used Bayesian quantile regression (BQR) methods. BQR was applied to stroke data collected between 2014 and 2018 in SA. The study considered lower, central and upper quantiles. Results: The study findings showed that stroke and modifiable risk factors were significantly associated with (p < 0.0001). The prevalence of stroke increased with cholesterol, hypertension, diabetes and heart-problem (OR 1.29, 1.33, 2.92 and 1.27) respectively. Modifiable and non-modifiable predictors had significant impact on stroke across quantiles. Conclusions: Most strokes were due to modifiable risk factors. Study findings showed significant impact of each predictor on stroke across quantiles.