In this paper we set out to examine the variation in CHD mortality by calendar year and birth cohort for all registered deaths in England and Wales between 1931 and 2005, a longer time period than previous studies.
Summary of main findings
We found evidence of a higher burden of CHD mortality for older age groups, which has only recently emerged in men, whereas it has been established in women for much longer. We also observed a previously reported peak  in the mortality rate among most age groups in the late seventies followed by a more recent decline.
We found a plateau in the CHD mortality rate among younger age groups for women, a trend which was not observed in men. Recent work has suggested that CHD mortality rates in younger men and women are heading for a plateau, based on an estimation of the average annual change in mortality rates since 1984, and probably reversing [10, 12]. The results presented here support this conclusion for young women (49 and younger), where the rate of change of CHD mortality rates appears to be converging on zero. The case in men is more complex: in younger men the rate of change of CHD mortality rates appears to have stabilized at a level below zero, suggesting that the current speed of decline has levelled out, but CHD mortality rates are still falling. In men between 45 and 54 years, there appears to be a small decrease which is less evident as in women. For older age groups of both men and women, the speed of reduction in CHD mortality rates continues to increase.
We observed that CHD mortality among younger age groups has increased in those born in the early twentieth century compared to those born in the late 19th century. This requires further study as the public health implications of a decline in survival from CHD in younger age groups may be stark. This pattern suggests that although significant advances have been made among older populations similar gains are not being made in those less than 60 years of age. There are a number of possible reasons for this including the targeting and efficacy of screening, the inclusion criteria for beginning of treatment regimes (which include age as a standard risk factor), and the current public health focus on mortality reduction in older populations. The trends in CHD mortality rates by age group and birth cohort show how more recent generations have enjoyed far lower CHD mortality rates than those born in the late 19th and early 20th centuries.
Strengths and Limitations
The paper presents age stratified CHD mortality rates over a long time period and contributes to the debate over the pattern of CHD mortality among younger age groups. It also provides an analysis of mortality trends across this time period within birth cohorts. It should be noted that comparisons between birth cohorts at the same age should be made with caution, as the numerators of the rates (number of CHD deaths) are based on slightly different definitions of CHD.
The reduction of the initial data to deaths from 1931 onwards is likely to have some effect on rates calculated for those born in 1895–1904 at ages 27–36. Age-stratified population levels for this cohort were calculated in a similar way as for other birth cohorts but there is likely to be an under-estimate of mortality rates in these groups. The possible under-estimate is in the order of 2% and should be borne in mind when interpreting results.
Any study using mortality data across multiple revisions of the International Statistical Classification of Diseases (ICD) will suffer attribution bias due to both the change between versions of ICD and the procedures to code deaths. In coding the data set within each ICD revision to a 'coronary heart disease' summary variable we have attempted to include underlying and contributing CHD mortality. In coding ICD-4, for example, angina pectoris both specified to CHD and without mention of CHD were coded to CHD in the parent data set (Table 1). This approach should be inclusive of all CHD over the time period. Differences across revisions of ICD also have the potential to affect the coding of CHD mortality. Deaths coded in versions of ICD prior to ICD-4 were considered so disparate as to be incomparable over time. For this reason the analysis here deals with deaths coded in 1931 onwards.
Jannsen and Kunst  examined the changes in deaths around ICD coding changes and found some evidence that ischaemic heart disease was affected. Their findings are difficult to interpret in relation to this paper as other major causes of mortality around changes in the coding such as floods or World Wars may also create unusual patterns in the data. Further, Jannsen and Kunst suggest that outliers may account for much of the observed differences in trend.
The change in ICD coding in the transition from ICD-9 to ICD-10 represents a large shift in disease coding  and for this reason comparisons between the last ten year period of deaths (1996–2005) and previous revisions should be interpreted with caution. In some other studies a "correction factor" is applied to adjust for differences in coding across different revisions of ICD. These correction factors, published by the UK Office for National Statistics among others, are useful for understanding trends around the transition period for ICD codes. They are not recommended for use in interpreting trend data over many revisions of ICD  and are not used here.
Changes in methods of death certification create a potential attribution bias that affects any study examining long term mortality using ICD coding. Coding of deaths may be effected by autopsy rates and the accuracy of CHD coding outside hospital. The 1995 United Kingdom Heart Attack Study suggested that up to age 65 death certification and coding were very accurate (within 4%), however, for deaths occurring above age 65, an over-estimate of about 20% was likely . Coupled with the continuous process of improving the accuracy of certification and coding over the last three decades, it is likely that the apparent fall in CHD death rates in older groups has been slightly over-estimated . While this adds potential variance to the mortality rates presented in this paper the falls in mortality rates themselves are real, just not quite as big in the elderly as perceived [15, 16].
Comparison with existing literature
Age-standardised CHD mortality rates for both men and women calculated using the coding frame described in Table 1 (results not shown here) show good agreement with Office for National Statistics (ONS) . It has been suggested that pre-1967 definitions of CHD were inconsistent due to the coding of some CHD deaths as 'other myocardial degeneration' (ICD codes 422 in ICD – 6 and ICD – 7). It has proved difficult to separate out other causes of death and previous work has used definitions which exclude code 422 in ICD-6 and ICD-7 . For this reason, the ONS results rely on two calculations for all pre-1967 rates, one including 'other myocardial degeneration' and one excluding it. The data presented here do not include 'other myocardial degeneration', and the 1950 to 1967 age-standardised mortality rates are very similar to those calculated by the ONS. A difficulty may arise when interpreting birth cohort results by age at death because the effect of revisions of ICD coding will affect different cohorts at different ages.
Previous work supports our finding that CHD mortality rates appear to be levelling out and perhaps reversing in younger age groups. Wilson and Siskind  studied death registrations for CHD using 5-year sex and age specific birth cohorts and found that in the youngest male cohorts (1950–54, deaths at ages 25 – 29) there was evidence of a flattening in rates. More recently a study set in the U.S.  described a reversal in previously declining CHD death rates. These authors identified an annual 1.3% increase in CHD mortality among women aged 35 to 44 between 1997 and 2002.