Methods
Directly age standardised (25–54, 55–64, 65–74 and 75–84 years) CHD and stroke mortality data for each year of the period from 1985 to 2010 were generated using CHD and stroke mortality data from the Public Health Information Systems database and the Northern Ireland Statistics and Research Agency for the ROI and NI, respectively. The following International Classification of Diseases (ICD) codes were used to identify deaths from CHD, coded as ischaemic heart disease: ICD9 (410–414) and ICD10 (I20–I25), and stroke: ICD9 (430–438) and ICD10 (I60–I69).
The 5-year age standardised (age groups as above, and overall 25–84 years) CHD and stroke mortality rates for 1985–1989 and 2006–2010 and percentage decrease in 5-year age standardised mortality rates from 1985–1989 to 2006–2010 were calculated for each country. Combined and gender specific analyses were performed.
Direct age standardised CHD and stroke mortality rates from 1985 to 2010 for each country were analysed using joinpoint regression analysis to identify points (years) where the slope of the linear trend changed significantly. This was performed separately for each of the age-standardised rates (25–54, 55–64, 65–74 and 75–84 years) and for men and women. The joinpoints (also referred to as turning points) are the calendar years at which the rate of change in CHD mortality changed significantly. Each joinpoint subdivides the time trend into distinct time periods, for example, if there is one joinpoint, there are two distinct time periods. The analysis begins with the assumption that there are no joinpoints, that is, the slope of the regression line fitted to the age standardised mortality rates does not change over the time period. It tests for at least one statistically significant joinpoint in the model. The joinpoint regression employed the permutation method. The model with the optimum number of joinpoints is selected by iteratively fitting models with no joinpoint up to a maximum of three joinpoints. We chose three joinpoints, or three changes in mortality, between 1985 and 2010 as visually this appeared to reflect the trends in CHD and stroke mortality data most accurately. The objective is to choose the model with the smallest number of joinpoints such that if an extra joinpoint is added the resulting improvement in the fit of the model is not statistically significant.
The pace of change in the CHD and stroke mortality rates was measured using annual percentage change (APC). The APC was computed for each distinct time period by fitting a regression line to the natural log of the age standardised mortality rates (response variable, y) and year (predictor variable, x) for each time period, that is, y = mx + c, where m is the slope, and c is the intercept of the regression line. The APC was then estimated using 100×(e–1) where e is the inverse of the natural log function. A 95% CI for the APC is also computed.
The analysis was performed using software developed by the Surveillance Research Programme of the US National Cancer Institute.