Smoking is one of the essential causes of stroke, cardiovascular disease and mortality. Different methods have been undertaken to investigate smoking status across health research (e.g., self-reported data and serum cotinine level). Studies have proved that self-reported data yields inaccurate measures. In serum cotinine level, there is no criterion to select the optimum cut-off point to identify the smokers and non-smokers. In addition, most of the studies assume linearity assumption while it might not exist. To overcome these problems, in this study, serum cotinine level is measured on a continuous scale and a flexible non-linear model is undertaken.

CRP, SC level, spline regression model, gender differences, NHANES dataset.