Response by Pase et al. to Letters by Archer and Chiu Regarding Article, “Sugar- and artificially sweetened beverages and the risks of incident stroke and dementia. A Prospective Cohort Study”

Our paper estimated beverage intake using data collected from a food frequency questionnaire (FFQ).¹ We adjusted our findings for energy intake derived from the FFQ. Dr. Archer questions the validity of this approach and quotes a group of eminent epidemiologists to support his stance.² The paper cited by Dr. Archer explains that the energy intake estimates in question should not be used as an exposure variable. However, these researchers also explicitly state that one should ‘use self-reported energy intake for energy adjustment of other self-reported dietary constituents to improve risk estimation in studies of diet-health associations.’² The improved validity after adjusting nutrient intakes for self-reported energy intake is clearly demonstrated in the OPEN study for FFQ-derived protein intake.³ As we used energy intake derived from the FFQ as a covariate and not as an exposure, our methods are actually in line with the recommended approach cited by Dr. Archer.

Dr. Archer also criticizes the validity of using a self-reported instrument to measure dietary intake more generally. He suggests that all self-report dietary data is inaccurate and not meaningful. This opinion is seeming extrapolated from the fact that energy intake alone is poorly estimated. Dr. Archer has been a vocal critic of self-report dietary instruments,^4–6 and his points have been rebutted extensively elsewhere.^{2, 7–9} Key criticisms of Dr. Archer’s position include ignoring a large body of science demonstrating successful uses of self-report dietary data and failing to acknowledge that self-report dietary data can still be valid and informative even if energy intake is poorly estimated.² We acknowledge that error occurs in all measurements, including in self-reported dietary assessment. However, as pointed out by others, such data can still be used to establish valid conclusions^{2, 10} and can set new questions for clinical trials.

We thank Dr. Chiu for his letter. Dr. Chiu asserts that, if results from our mediation analysis had been included in the main results of the study, the authors might not have concluded that artificially sweetened beverages (ASB) increase the risk of stroke and dementia. We would like to clarify that, whereas we described an association between a higher intake of ASBs and an increased risk of stroke and dementia, we did not conclude that the association was causal. Indeed, our concluding paragraph calls for other studies to replicate our findings and to determine the underlying mechanisms.

Although we performed mediation analysis, this does not provide definitive answers as to the mechanisms linking ASB to stroke and dementia risk. For example, the association between ASB and all-stroke was mediated by prevalent hypertension, whereas the association with ischemic stroke was not. Moreover, the association between ASB and dementia remained after the exclusion of subjects with prevalent diabetes. Without knowing and tracking dietary habits and vascular risk factor burden from an earlier age, it is difficult to tease apart the temporal sequence between high intake of ASBs, development of a high vascular risk factor burden (or vise versa), and incident events such as stroke. Our study was upfront about the limitations of our observational design. We explicitly acknowledged that we were unable to determine whether ASB intake increased the risk of incident stroke and dementia through risk factors such as diabetes mellitus or whether people with a greater risk factor burden were simply more likely to consume diet beverages.

With respect to our classification and analysis of beverage intake, we stratified participants by intake of sugar-sweetened beverages (SSB) and ASB independently. It was necessary to use different intake thresholds to define each variable owing to their relative distributions. Consequently, the hazard ratios for SSB and ASB are not directly comparable. Therefore, it is uninformative to compare incident rates between levels of ASB and SSB. To avoid limiting our sample size, we chose not to examine subjects who consumed only one type of beverage exclusively. The fact that SSB intake was neither associated with incident stroke nor dementia suggests that the association between ASB and these same outcomes was not mediated by shared intake of SSB consumption.