calculated for each of 2 periods (ie, 2000-2003 and 2013-2016). These 4-year periods allowed for comparison of age-specific rates between the most recent vs the oldest years within the study period while avoiding instability of data from potentially outlying individual years. Second, we used the age-period-cohort analysis webtool developed by Rosenberg et al17 to identify APCs in mortality rates during the study period for individual ages. Age-period-cohort models separate trends into age associations, which reflect natural history; period associations, which may reflect changes in the environment that affect all age groups; and cohort associations, which reflect differences in risk across birth cohorts. The age-period-cohort analysis webtool implements a panel of estimable age-period-cohort functions and corresponding Wald tests in R code following the input of age-specific numbers of events and person-years over time, which were uploaded to the tool in a comma-separated values format. For this analysis, we limited the cohort to the population aged 20 to 80 years because of low case numbers and statistical instability among age groups outside this range.
