We found that rates of CHD events calculated from cohort risk factor information were lower than the observed rates in most of the source communities. However, the slopes of the observed and expected CHD rates were similar over time. Further, CHD death was well predicted in most of the communities. These results suggest that traditional individual level risk factor information from a cohort can provide predicted CHD event and CHD death rates that are useful indicators of trends expected from community surveillance but that prediction of absolute rates is less reliable.
These results confirm previous studies showing that aggregate risk factor changes track with expected CHD mortality and mortality changes.(4
) Since a single model of event rates was used in our study across visits and communities, differences in the expected rates over time and across communities are driven by differences in the risk factor distributions. Our findings also support Ford and colleagues suggestion that approximately half the decline in CHD death was due to changing risk factors.(3
) However, for all CHD events, the estimated decline in events was greater than the observed decline, suggesting that risk factor changes may be overly optimistic in predicting CHD morbidity change.
Our results add to the previous literature in examining variability across populations as well. While previous studies have found weak associations of individual baseline risk factors with CHD mortality (15
) we found no evidence that the expected CHD mortality based on combined risk factor levels was different from the observed mortality in four out of our five groups. However, we did see substantial evidence that the expected combined CHD events were different from the observed in four of five groups.
The findings build on the combination of strengths in the data, including the ability to compare across four different geographic areas as well as two racial groups. The availability of multiple sources of prospective data collection, including adjudicated surveillance of fatal and non-fatal CHD events for the entire community, allowed us to examine all CHD events rather than focusing primarily on mortality, while the collection of risk factor data at multiple time points allowed for flexibility in the modeling strategies used. Using the combined cohort and surveillance offers additional information over an ecological analysis in which the relationship between the risk factors and outcomes is unknown on an individual level, as well as additional information over only a cohort design in which the relationship of the cohort to the larger source population is unknown.
This study makes two main assumptions. First, it assumes that the risk factor distribution in the community is the same as the distribution in the cohort, conditioned on age, gender and ethnic distribution. This assumption, while strong, is not unreasonable. The cohort was chosen and recruited to represent the community at the time of enrollment and response rate for the clinic visit varied from 67% of identified eligible participants in Minnesota to 46% in Jackson.(17
) While this response rate is within the range reported in similar multi-center studies (4
), participants were more likely to have completed high school, have a higher income, not currently smoke, report excellent health (white men and women and black men), have hypertension (black men) and have high cholesterol.(17
) However, the mean total cholesterol levels in the black cohort members somewhat higher than would be expected relative to the white cohort members based on patterns seen in national cross-sectional surveys (1
). This relatively healthy cohort may explain why observed rates of MI were higher than expected at the beginning of the study period. However the persistence of the difference between expected and observed is perhaps less expected, as the healthy cohort effect might have been expected to diminish over time. Retention rates in the cohort for each of the four visits ranged from 93 to 81% and overall five year migration rates were low (5 to11% in 1990 and 1 to 9% in 2000). Aside from the Forsyth County black community, the consistency of the change over time for expected and observed rates suggests that the migration patterns were not a strong influence on these results.
The second assumption is that the relationships between the risk factors and the likelihood of a CHD event observed in the cohort are the same in the reference community cohort. While it is highly plausible that the total relationships are not identical, the assumption of some biologically consistent component seems reasonable. This assumption of biological consistency is used in the wide application of the Framingham risk score (18
) and other risk scores in clinical practice and has been confirmed in studies which compare the relative hazard for each risk factor across populations.(16
) Participation in the study itself might also dilute the effect of risk factors, as participants might be more likely to get treatment for elevated risk factors than non-participants.
In conclusion, total CHD incidence declined by 1 to 3% per year and CHD mortality declined by 3 to 6% per year over the study period among 53–64 year old men and women in all ARIC communities except the Forsythe County black community, where the CHD incidence increased. CHD risk factor trends within each community predicted a decline of 2 to 3% per year in both CHD incidence and mortality. Overall, CHD rates parallel the trends in expected rates based on cohort risk factors but declines in mortality were steeper than expected, particularly among blacks.