Statistics analysis

The enumeration data including demographic characteristics were expressed as frequency and rate, while measurement data were expressed as X¯±S, median and range. Longitudinal data were analysed using a linear mixed model (LMM), reporting the β coefficient and 95% confidence intervals (95% CI). LMM was used to assess the change in medication rate over the follow-up and to evaluate which variables were associated with the trend of change. The medication rate in each follow-up was used as the outcome variable to fit the null model and the random intercept-slope model. After the addition of the level 1 explanatory variable (scores of the barriers and benefits perceived during medication) and the level 2 explanatory variable (demographic characteristics), the final model was fitted. The modelling of the linear mixed models was conducted using the Proc Mixed of SAS software. All statistical analyses were performed using SAS 9.4 software and SPSS 22.0 software. P < 0.05 was considered to have statistical significance.