Quantifying the differential relationship between red blood cell docosahexaenoic acid and cognition based on APOE-e4 carriership in multiple cohorts: illustration of an emerging statistical approach

<h2>Abstract</h2> Evidence is mounting that red blood cell membrane levels of docosahexaeonic acid (DHA) are directly associated with improved cognitive performance, with emerging questions about whether this relationship may be stronger in some individuals vs. others, especially those at high genetic risk of cognitive impairment via the <i>APOLIPOPROTEIN-E</i> (<i>APOE</i>) <i>ε4</i> allele. We investigated this relationship in <i>n</i> = 2158 participants without dementia from four separate cohorts (Offspring, Gen3, Omni I and Omni II) of the multi-cohort Framingham Heart Study, aged 50 to 70 years old. To demonstrate an emerging alternative of improved data sharing, we conducted the analysis using only cohort-level, summary data which is non-individually identifiable and easily shareable. We detected a statistically significant interaction between <i>APOE-ε4</i> status and DHA in multivariable models adjusting for relevant demographic and medical history covariates and predicting cognitive impairment (<i>p</i> = 0.039). Stratified adjusted model results detected a stronger inverse association between cognitive impairment and DHA for <i>APOE-ε4</i> carriers (-0.13; 95 % CI: (-0.20, -0.06) minutes per 1 % increase in DHA), as compared to non-carriers (-0.06; 95 % CI: (-0.11, -0.02)), though both were statistically significant (<i>p</i> < 0.05). We also analyzed the data using two, standard, widely used statistical approaches (1) Individual level data analysis (identical results) and (2) standard meta-analysis (slightly weaker evidence of interaction; <i>p</i> = 0.047). Our analysis demonstrates not only the putative importance of DHA on cognition, especially among <i>APOE-ε4</i> carriers, but the potential for an emerging class of data-sharing methods to accelerate the future pace of cross-cohort analyses by not requiring access to individual-level data.