Prevalence estimates and standard errors for drinking status, binge drinking, and average volume were computed for adults ages 60+ across survey years. We used joinpoint regression by applying the National Cancer Institute Joinpoint Regression Program desktop version 4.3.1.0 (https://surveillance.cancer.gov/joinpoint/) to examine possible nonlinear trends by fitting the simplest joinpoint regression to the natural logarithm of estimates, weighted by the square of the estimate divided by variance at each year, and back transformed to obtain the slopes on the prevalence scale. Selection of joinpoints was based on the permutation test at an overall significance level of 0.05 (Kim et al., 2000). The level of statistical significance (p-value) based on the permutation test does not account for correlation across years of survey due to the same primary sampling units that were used between years (Botman et al., 2000; Parsons et al., 2014). Therefore, the p-values should be considered as approximately correct. The slope estimates from log-linear segments were used to estimate annual percent changes (APC) and average annual percent changes (AAPC). APC was presented for each line segment connected at the joinpoint(s).
