In logistic models, we used a propensity score as a covariate to control potential confounding factors (viz., gender, age, education level, monthly allowance, insurance, and hometown) as described previously.5,6 In brief, propensity score for each participant was calculated through logistic models, where PKA was used as dependent variable with the confounding factors mentioned above as independent variables, generating a score which represented the possibility of having PKA for each participant