We were not surprised to see that crude results significantly differ from the fully adjusted results. This was the main hypothesis of our study, and the demonstration of how the direction of the association changed after we considered both traditional risk factors for heart disease and indicators of poor health status is the principal finding. In observational studies, crude analyses can lead to incorrect conclusions in the presence of important confounding. As shown in Table 1 in our report, persons with the lowest cholesterol levels had a greater prevalence of several indicators of poor health and more cardiovascular risk factors than did those with higher cholesterol levels.

In addition, any assessment of risk requires comparison with a reference group. When the reference group of persons with lower cholesterol levels in an older population includes those with both life-long low cholesterol levels and declining low cholesterol levels caused by illness, the reference group will have a higher mortality rate and the high-cholesterol group will not seem to be at increased risk. We believe that illustrating the heterogeneity of the older population and providing evidence that comorbidity and frailty can "hide" the association between cholesterol levels and death is an important contribution to the field of cardiovascular research.

It is worth noting that in our study, elevated cholesterol levels ( 6.20 mmol/L) were already significantly associated with increased risk for death from coronary heart disease when all follow-up deaths were considered (relative risk, 1.45; P < 0.05 [Table 2 and the third panel of Figure 1 in our report]). We do not exclude the data in the first year but emphasize the different findings in the data after the first year because of the likelihood that the 1-year results are biased as a result of unmeasured confounding by comorbid conditions. Separate consideration of deaths in the first year and those occurring thereafter constitutes not adjusting twice but additional evaluation of the effects of residual confounding.

The issues related to the 4S study [1] have already been addressed by Gaziano and colleagues [2] in response to similar concerns expressed in a previous letter by Dr. Grossman. The discussion about the 4S study findings, therefore, does not have to be reiterated.

With regard to the West of Scotland [3] and CARE [4] studies, it is important to emphasize the study designs. In the former study, the primary end points were the incidence of nonfatal myocardial events and the rate of death from coronary heart disease. In the latter study, the end points were fatal and nonfatal coronary events and stroke. At the end of the study period, both trials succeeded in demonstrating that the pravastatin group had significantly lower event rates than the placebo group. It would have been unethical to prolong the trials after the major end points were reached.

The totality of evidence points toward a cause and effect association between elevated cholesterol levels and coronary heart disease. We believe that our study provides solid and statistically sound observational evidence that this association persists into old age.