Reply to: COMMENT ON: SERUM CHOLESTEROL AND INCIDENT ALZHEIMER’S DISEASE: FINDINGS FROM THE ADULTS CHANGES IN THOUGHT STUDY

To the Editor:

We thank Ilkin and Tasci for their comment on our article evaluating associations between high-density lipoprotein cholesterol (HDL) and non-HDL-C levels at specific ages and subsequent Alzheimer’s disease (AD) risk.¹ Ilkin and Tasci highlight the potential for unmeasured confounding in our observational analysis, especially as it relates to the role of physical activity.² We agree that adjusting for a validated measure of physical activity would have been ideal. Unfortunately, given our analytical design, the cohort-based interview question on physical activity available in the ACT study was unavailable for many participants during the age band exposure window of interest because the window predated the individual’s enrollment in ACT.

We estimated a U-shaped relationship between non-HDL-C and AD risk. Given the hypothesized direction of association between physical activity and AD and physical activity and non-HDL-C levels, we would anticipate potentially different impacts of the unmeasured confounder, physical activity, on our findings at the two ends of the non-HDL-C spectrum. Assuming that less physical activity increases risk of AD and that less physical activity increases non-HDL-C cholesterol levels, the modest elevated AD risk we found with high non-HDL-C levels could potentially be explained by residual confounding. One way to estimate the robustness of observational research results is to use the E-value.³

According to Haneuse et al, calculating an E-value answers the question, “how strong would the unmeasured confounding have to be to negate the observed results?”⁴ For instance, we found that in people aged 60 to 69, those with an average non-HDL-C level of 210 mg/dL had a 16% greater AD hazard (HR=1.16, 95% CI=1.01–1.33) than those with an average of 160 mg/dL. The E-value for this HR point estimate is 1.45 meaning that residual confounding could explain away this estimated association if there exists an unmeasured covariate having a relative risk association at least as large as 1.45 with both AD and with high non-HDL-C levels. In our study, the HRs for some of the known, potentially strong AD risk factors were 1.44 (95% CI, 1.17–1.77) for lower educational attainment (high school or less compared to at least some college), 1.16 (95% CI, 0.95–1.42) for ever-treated hypertension, and 1.15 (95% CI, 0.76–1.75) for ever-treated diabetes. Further, a prior ACT study⁴ reported an association between lack of regular exercise and AD of magnitude HR=1/0.69=1.45, which is the same order of magnitude as education in our analysis. However, that study was restricted to original cohort participants in the top three quartiles of cognition at baseline (based on CASI) and was focused on education as reported at that time (mean age at baseline, 74 years). Given the above, we agree the magnitude of the association we estimated between high non-HDL-C levels and increased AD risk would likely be attenuated had we been able to include adjustment for physical activity levels, and that the attenuation might plausibly have been sufficient to explain away that estimate or at least result in a confidence interval overlapping the null (especially as the E-value for the confidence interval was 1.09).

Conversely, we found that in people aged 60 to 69, those with an average non-HDL-C level of 120 mg/dL had a 29% greater AD hazard (HR=1.29, 95% CI=1.04–1.61) than those with an average of 160 mg/dL. We do not think unmeasured confounding from physical activity would explain this finding of elevated AD risk with low non-HDL-C. In fact, given the hypothesized direction of association between physical activity and AD risk and non-HDL-C levels, we anticipate having estimated an association of stronger magnitude between low non-HDL-C and increased AD had we been able to adjust for physical activity. We appreciate the opportunity to further explore this robustness check in response to the comment.