In reply
We thank Drs Connick and Stacpoole for their interest in our study. In our analyses we found a paradoxical association between the LDL-C levels at baseline and subsequent ischemic stroke risk such that elevated levels were associated with a decreased risk of the outcome. None of the variables in our final models violated the Cox proportional hazard assumption, suggesting no age-dependent effects. We believe that our findings may reflect temporal trends of increased statin use among our participants, and we tried to investigate this hypothesis using 2 sets of analyses. In the first set of analyses, we found that treatment with cholesterol-lowering medications at any time during follow-up modified the effect of LDL-C on stroke risk. Participants who were receiving treatment had a lower risk of ischemic stroke than those who were not receiving treatment, supporting the protective effect of cholesterol-lowering medications. The lack of effect of LDL-C on stroke risk among those not treated with medication could be owing to lack of power, low numbers of atherosclerotic strokes for which LDL-C may be pathological, or a true lack of effect. To further explain our results, we carried out exploratory analyses using imputed values drawn from a 10% sample of the Northern Manhattan Study that had annual laboratory measurements. To account for pharmacological effects on changes in LDL-C and subsequent risk of stroke, we included cholesterol-lowering medications and an interaction with time as a covariate in the mixed-effects models in Table 5. Using the imputed values, we found that LDL-C as a categorical variable was not associated with a decreased risk of ischemic stroke; we also found similar results including LDL-C as a continuous measure but did not report it owing to space limitations (adjusted hazard ratio per 10 mg/dL increase in LDL-C, 1.40; 95% confidence interval, 1.16-1.69). Because we found nonparadoxical results for LDL-C as a continuous variable, no further investigation was necessary to test potential nonlinearity of LDL-C. These results should be interpreted with caution, however, as our analyses were exploratory and based on a small sample with repeated measures. We believe that these results support, in part, our hypothesis that the true association between LDL-C and ischemic stroke is not paradoxical. Given the limited proportion of participants for whom we have repeated measures, the conclusion that there is a positive association cannot be inferred from our data. We believe, moreover, that carrying out further analyses based on this small sample may lead to overinterpretation of a limited set of data. Another study with repeated measures of LDL-C that does not use imputed values may answer the question of the effect modification of age and whether there is a linear association with lipid panel components and risk of ischemic stroke. We agree with the authors that it may be that LDL-C closer to the time of stroke may be more crucial, as manifested by the findings of the case-control Northern Manhattan Stroke Study.1